WEBVTT

00:00:04.189 --> 00:00:06.809
Welcome to Artificially Intelligent Marketing,

00:00:07.030 --> 00:00:09.949
a weekly podcast where we stay on top of the

00:00:09.949 --> 00:00:13.250
latest trends, tips, and tools in the world of

00:00:13.250 --> 00:00:16.070
marketing AI, helping you get the best results

00:00:16.070 --> 00:00:19.210
from your marketing efforts. Now let's join our

00:00:19.210 --> 00:00:24.120
hosts, Paul Avery and Martin Broadhurst. And

00:00:24.120 --> 00:00:26.239
welcome again to Artificially Intelligent Marketing,

00:00:26.379 --> 00:00:30.600
episode 52, but the first episode of 2026. Paul

00:00:30.600 --> 00:00:33.600
Avery here, joined as always by Martin Broadhurst.

00:00:33.719 --> 00:00:36.159
Happy New Year, Martin. How are you doing? Great,

00:00:36.259 --> 00:00:38.659
Paul. And to be perfectly honest, I'm not even

00:00:38.659 --> 00:00:41.539
sure if it is actually me here or if it's one

00:00:41.539 --> 00:00:44.640
of the many agents or sub -agents that I have

00:00:44.640 --> 00:00:47.259
operating for me and on behalf of me at any given

00:00:47.259 --> 00:00:51.119
time at the moment. It's been a busy start to

00:00:51.119 --> 00:00:54.039
the new year, testing out all of these new AI

00:00:54.039 --> 00:00:58.079
tools, more of which later. Well, as far as I'm

00:00:58.079 --> 00:01:00.939
concerned, across the camera here, you are oozing

00:01:00.939 --> 00:01:03.280
charisma. So I know for sure it's one of your

00:01:03.280 --> 00:01:08.840
agents. Well, it's the start of 2026. Of course,

00:01:08.859 --> 00:01:11.480
we have the obligatory Friends and Predictions

00:01:11.480 --> 00:01:14.260
episode. So for all those people who've probably

00:01:14.260 --> 00:01:16.120
already listened to 50 of these of different

00:01:16.120 --> 00:01:19.579
types, here's another one. We've got some interesting

00:01:19.579 --> 00:01:22.480
takes, potentially a few contrarian takes in

00:01:22.480 --> 00:01:24.439
here as well. So I think it's hopefully still

00:01:24.439 --> 00:01:27.140
going to be worth your time, dear listener. So

00:01:27.140 --> 00:01:29.219
why don't we just get straight on into it then?

00:01:29.500 --> 00:01:32.900
What do we think AI is going to look like for

00:01:32.900 --> 00:01:35.760
marketers by the end of 2026? What's your first

00:01:35.760 --> 00:01:40.319
prediction, mine? So mine is that the small AI,

00:01:40.519 --> 00:01:42.379
scrappy little startup, you might have heard

00:01:42.379 --> 00:01:50.140
of them, they're known as Google. of chrome with

00:01:50.140 --> 00:01:56.599
a fully agentic browser experience akin to perplexity's

00:01:56.599 --> 00:02:01.299
comet or flawed for chrome i think this one is

00:02:01.299 --> 00:02:06.519
just so obvious the way that the whole market

00:02:06.519 --> 00:02:10.740
is moving they don't have one at the moment but

00:02:10.740 --> 00:02:14.639
the gemini platform could so easily Jump right

00:02:14.639 --> 00:02:16.560
in there. And to be honest, having been using

00:02:16.560 --> 00:02:18.979
these tools for a while now, both Perplexity's

00:02:18.979 --> 00:02:23.819
Comet and Claude for Chrome, which I've been

00:02:23.819 --> 00:02:25.580
playing around with for the past couple of weeks,

00:02:25.740 --> 00:02:30.039
I just think that it's a must -have, essential

00:02:30.039 --> 00:02:34.979
add -on to the browser experience. And for marketers,

00:02:35.319 --> 00:02:38.680
well, let me tell you, if you can get to grips

00:02:38.680 --> 00:02:41.479
with understanding how to prompt and instruct

00:02:41.479 --> 00:02:44.990
these agentic browsers, there is a massive unlock

00:02:44.990 --> 00:02:47.729
for individual productivity to be had. Yeah,

00:02:47.770 --> 00:02:50.430
I'm pretty excited about Google having one specifically,

00:02:50.669 --> 00:02:54.870
just because I would expect the capabilities

00:02:54.870 --> 00:02:58.289
in a bunch of work -driven platforms to be way

00:02:58.289 --> 00:03:01.370
better. You know, we're a Google Apps organization,

00:03:01.650 --> 00:03:05.310
so Gmail, Google Drive, Google Slides, Google

00:03:05.310 --> 00:03:08.289
Docs, you know, the ability to say, go search.

00:03:08.889 --> 00:03:11.569
a bunch of these files and then write something

00:03:11.569 --> 00:03:14.409
in this document and then update this spreadsheet

00:03:14.409 --> 00:03:17.810
and then mark the trello card done like that

00:03:17.810 --> 00:03:20.110
would be really quite nice if we could get some

00:03:20.110 --> 00:03:24.169
level of agentic behavior built into chrome knows

00:03:24.169 --> 00:03:26.150
a lot about us already whether we like it or

00:03:26.150 --> 00:03:29.069
not in terms of the websites we visit and should

00:03:29.069 --> 00:03:30.990
be logged into a bunch of stuff that we want

00:03:30.990 --> 00:03:33.110
to be able to do things in automatically and

00:03:33.110 --> 00:03:35.849
yeah it feels like a no -brainer but They've

00:03:35.849 --> 00:03:38.389
been a bit notoriously slow. I mean, if I had

00:03:38.389 --> 00:03:41.050
to guess, I'd say this capability exists within

00:03:41.050 --> 00:03:43.650
Google already and it just hasn't been launched.

00:03:43.750 --> 00:03:47.289
So why the hold up? Yeah, I would agree. They're

00:03:47.289 --> 00:03:49.389
probably testing this extensively internally

00:03:49.389 --> 00:03:53.810
because they want to get it right. Every new

00:03:53.810 --> 00:03:56.689
product release that comes out of Google is scrutinized

00:03:56.689 --> 00:04:01.030
probably more heavily and more closely than competitors.

00:04:02.209 --> 00:04:05.330
perplexity can throw out comet and people go

00:04:05.330 --> 00:04:08.469
oh that's cool and it does some tasks over a

00:04:08.469 --> 00:04:11.689
short time horizon but if google were to launch

00:04:11.689 --> 00:04:13.650
a similar product that had a similar short time

00:04:13.650 --> 00:04:17.009
horizon people might go oh it's a bit crap so

00:04:17.009 --> 00:04:18.670
i think they'll probably want to get it right

00:04:18.670 --> 00:04:22.689
to the point where it can run and execute tasks

00:04:22.689 --> 00:04:25.970
over the long term and maybe manage sub -agents

00:04:25.970 --> 00:04:28.970
so they can show something that's like really

00:04:28.970 --> 00:04:33.000
impressive and makes people go oh yeah this is

00:04:33.000 --> 00:04:34.779
this is special and i think you're right the

00:04:34.779 --> 00:04:36.480
integration that they've got with the google

00:04:36.480 --> 00:04:40.560
workspace environment the authentication that

00:04:40.560 --> 00:04:44.259
they have through google authentication tools

00:04:44.259 --> 00:04:51.180
and just the vast amount of user data that they'll

00:04:51.180 --> 00:04:55.139
have on behavior will make their agent a step

00:04:55.139 --> 00:04:58.240
above the competition i dare say and i say that

00:04:58.240 --> 00:05:02.000
as someone who has started using claude for chrome

00:05:02.000 --> 00:05:05.800
and been really impressed with its abilities

00:05:05.800 --> 00:05:10.300
and can see that a native gemini one in chrome

00:05:10.300 --> 00:05:14.240
well it would be a true disruptor in the browser

00:05:14.240 --> 00:05:16.300
field for sure i guess it'd be quite easy to

00:05:16.300 --> 00:05:19.019
know if this is a done or not done eos style

00:05:19.019 --> 00:05:21.899
uh in terms of whether we got this prediction

00:05:21.899 --> 00:05:24.139
right or not because obviously it will get launched

00:05:24.139 --> 00:05:26.759
but what do you think are like The practical

00:05:26.759 --> 00:05:29.180
ramifications for marketers here, is there anything

00:05:29.180 --> 00:05:31.220
that we should be thinking about or doing in

00:05:31.220 --> 00:05:33.779
the lead up to this potential launch? Yeah, I

00:05:33.779 --> 00:05:37.759
think if people can get their hands on something

00:05:37.759 --> 00:05:43.019
like Claude or Chrome or have a play with Perplexity's

00:05:43.019 --> 00:05:48.680
comic browser, just get used to briefing in these

00:05:48.680 --> 00:05:51.759
agents. It's really interesting to see them operate

00:05:51.759 --> 00:05:55.319
and how they get stuck. They work really well.

00:05:55.720 --> 00:05:59.879
with very well -specified tasks, even to the

00:05:59.879 --> 00:06:03.240
point of sometimes saying, then click on this

00:06:03.240 --> 00:06:07.699
button. So directing them through the UI. And

00:06:07.699 --> 00:06:10.000
what's interesting is I can see a world where

00:06:10.000 --> 00:06:13.279
it's almost like the skills .md, the skills markdown

00:06:13.279 --> 00:06:15.720
file that they recommend using for cloud code

00:06:15.720 --> 00:06:20.040
agents now. The more granularity and specificity

00:06:20.040 --> 00:06:22.379
that you put in a task, the more repeatable and

00:06:22.379 --> 00:06:25.029
predictable it's going to be. And I can easily

00:06:25.029 --> 00:06:28.850
see marketers that are regularly pulling stats

00:06:28.850 --> 00:06:31.610
from, you know, maybe it's an SEO tool and you

00:06:31.610 --> 00:06:33.730
want to pull keyword ranking reports or you're

00:06:33.730 --> 00:06:36.629
doing Google Analytics reporting and moving data

00:06:36.629 --> 00:06:40.069
from here to there. Having the steps of that

00:06:40.069 --> 00:06:42.649
task really well documented and then sticking

00:06:42.649 --> 00:06:46.189
that into one of these browser agents, that's

00:06:46.189 --> 00:06:48.209
going to be the thing that makes them go from.

00:06:48.899 --> 00:06:51.139
interesting tool but makes a lot of errors to

00:06:51.139 --> 00:06:53.279
oh actually this is something i'm going to use

00:06:53.279 --> 00:06:57.519
on a weekly daily even basis yeah i think that's

00:06:57.519 --> 00:06:59.279
a really good suggestion i think the other thing

00:06:59.279 --> 00:07:03.000
when this comes as we expect it to to google's

00:07:03.000 --> 00:07:05.759
product suite It's going to open up access to

00:07:05.759 --> 00:07:07.759
people, especially in a work environment that

00:07:07.759 --> 00:07:10.860
might not have the freedom and flexibility to

00:07:10.860 --> 00:07:12.939
use whatever tool they want to have a play around

00:07:12.939 --> 00:07:15.319
with. Maybe, you know, you can still follow Martin's

00:07:15.319 --> 00:07:16.759
advice because there's nothing to stop you doing

00:07:16.759 --> 00:07:18.759
this on your own personal computer and getting

00:07:18.759 --> 00:07:20.600
to grips with these tools, but actually using

00:07:20.600 --> 00:07:22.339
them for work. You might be in an environment

00:07:22.339 --> 00:07:26.180
where everything's Google based. And if Gemini

00:07:26.180 --> 00:07:28.199
doesn't do it, then I'm afraid you can't use

00:07:28.199 --> 00:07:30.100
it. Whereas obviously that would open this access

00:07:30.100 --> 00:07:32.379
up. I guess by extension, we can probably expect

00:07:32.379 --> 00:07:36.410
to see. co -pilots maturation within edge and

00:07:36.410 --> 00:07:38.569
probably similar capabilities for people who

00:07:38.569 --> 00:07:40.829
are tied to the microsoft system yeah i think

00:07:40.829 --> 00:07:43.610
they've more or less confirmed that these kinds

00:07:43.610 --> 00:07:46.810
of abilities are coming to edge and are being

00:07:46.810 --> 00:07:49.529
rolled out slowly in the coming months i'd expect

00:07:49.529 --> 00:07:53.410
that probably to be in most people's browsers

00:07:53.410 --> 00:07:58.449
by q2 but it'll be interesting to see how how

00:07:58.449 --> 00:08:01.339
much of this ability they give people and in

00:08:01.339 --> 00:08:06.139
the copilot 365 environment do the users with

00:08:06.139 --> 00:08:09.399
copilot 365 chat which is the free version of

00:08:09.399 --> 00:08:12.420
copilot do they get access to this or is it only

00:08:12.420 --> 00:08:16.180
for people that have the 365 pro license as well

00:08:16.180 --> 00:08:18.920
that'll be an interesting question because certainly

00:08:18.920 --> 00:08:22.360
if you're using low cost or free based versions

00:08:22.360 --> 00:08:24.759
of these tools once things get agentic they eat

00:08:24.759 --> 00:08:27.870
tokens for breakfast and lunch and dinner and

00:08:27.870 --> 00:08:29.689
they snack all day and they eat millions of tokens

00:08:29.689 --> 00:08:34.070
um makes it much harder to deliver that stuff

00:08:34.070 --> 00:08:36.029
for free although i've got some ideas and some

00:08:36.029 --> 00:08:38.110
of the trends and predictions i'm going to make

00:08:38.110 --> 00:08:40.549
as to how they might still make money from from

00:08:40.549 --> 00:08:42.929
those things but um nice one martin that's a

00:08:42.929 --> 00:08:46.289
great first prediction martin and i kind of did

00:08:46.289 --> 00:08:48.129
this independently so some of our predictions

00:08:48.129 --> 00:08:50.129
overlap a bit but actually they're surprisingly

00:08:50.970 --> 00:08:53.169
not different, but take different angles. So

00:08:53.169 --> 00:08:56.549
my first prediction is actually about generative

00:08:56.549 --> 00:08:59.309
engine optimization, answer engine optimization,

00:08:59.549 --> 00:09:02.049
whatever you want to call it. I think we're going

00:09:02.049 --> 00:09:04.990
to see, I think we think we've seen the explosion

00:09:04.990 --> 00:09:07.450
in people trying to get good at this. And I don't

00:09:07.450 --> 00:09:11.350
think people have even begun. Now, I've been

00:09:11.350 --> 00:09:14.549
somewhat critical of companies piling on the

00:09:14.549 --> 00:09:20.379
AEO, GEO bus. I think people are right. They're

00:09:20.379 --> 00:09:23.399
just a bit too early and they're probably not

00:09:23.399 --> 00:09:26.240
doing it the right way, which is I do think we're

00:09:26.240 --> 00:09:28.480
going to need this. We do want to come up in

00:09:28.480 --> 00:09:32.240
AI answers, but whether or not there's some crazy,

00:09:32.379 --> 00:09:36.259
amazing AEO, GEO specific tactics that you need

00:09:36.259 --> 00:09:39.460
to deploy or rather you just be really good at

00:09:39.460 --> 00:09:41.580
what was good for SEO and a lot of it ends up

00:09:41.580 --> 00:09:43.139
being good for AEO is maybe something we can

00:09:43.139 --> 00:09:45.519
discuss in a bit. But I was actually on a LinkedIn

00:09:45.519 --> 00:09:47.700
live. The reason I put this one in is I was on

00:09:47.700 --> 00:09:51.110
a LinkedIn live yesterday. And we had our head

00:09:51.110 --> 00:09:54.470
of paid ads as part of Supreme Optimization,

00:09:54.549 --> 00:09:56.809
which is the paid ads specialist within Supreme

00:09:56.809 --> 00:10:00.269
Group. And she was telling us how we haven't

00:10:00.269 --> 00:10:04.389
seen any reduction in Google ad clicks and spent.

00:10:04.590 --> 00:10:06.750
And we had a really good discussion. And the

00:10:06.750 --> 00:10:09.529
conclusion that we came to is that at the current

00:10:09.529 --> 00:10:13.250
point, people are maybe using AI to better understand

00:10:13.250 --> 00:10:16.710
the product landscape, self -educate. learn at

00:10:16.710 --> 00:10:18.250
the different options available but then they

00:10:18.250 --> 00:10:20.929
still google the product category or the product

00:10:20.929 --> 00:10:23.850
name and then they go and visit the website to

00:10:23.850 --> 00:10:25.710
read about the product and then they buy it there

00:10:25.710 --> 00:10:27.990
which has all sorts of attribution nightmares

00:10:27.990 --> 00:10:33.129
associated with it to be honest but what i predict

00:10:33.129 --> 00:10:38.309
over the course of 2026 is more and more people

00:10:38.309 --> 00:10:41.789
turn to ai we it maybe gets slightly better at

00:10:41.789 --> 00:10:44.620
driving traffic to product pages and so maybe

00:10:44.620 --> 00:10:46.320
we can track that a bit more so i think most

00:10:46.320 --> 00:10:49.799
people will start to see their traffic directly

00:10:49.799 --> 00:10:54.960
from perplexity chat gpt gemini increasing but

00:10:54.960 --> 00:10:58.340
i also think we'll start to see the ability to

00:10:58.340 --> 00:11:02.840
purchase in these chat bots as well which i think

00:11:02.840 --> 00:11:06.860
will be where aeo geo starts to become really

00:11:06.860 --> 00:11:10.399
important right because you're missing out on

00:11:10.399 --> 00:11:13.740
purchases if you're not in the answers and i

00:11:13.740 --> 00:11:15.840
want to talk about that as as part of a separate

00:11:15.840 --> 00:11:18.820
prediction maybe in a little bit but um what

00:11:18.820 --> 00:11:21.980
i think that means is for marketers we're probably

00:11:21.980 --> 00:11:25.139
going to be shifting towards the types of top

00:11:25.139 --> 00:11:27.419
of the funnel content that's like comparisons

00:11:27.419 --> 00:11:30.000
or the best x for y some of this we've talked

00:11:30.000 --> 00:11:32.559
about already on the podcast martin things that

00:11:32.559 --> 00:11:36.960
aren't a common problems um that by personas

00:11:36.960 --> 00:11:39.159
might be facing and again anybody who's done

00:11:39.159 --> 00:11:40.860
inbound marketing over the last 10 years this

00:11:40.860 --> 00:11:43.120
probably sounds quite familiar right so if you're

00:11:43.120 --> 00:11:44.899
already doing this stuff you're probably already

00:11:44.899 --> 00:11:49.559
doing quite well in an aeo and geo and then i

00:11:49.559 --> 00:11:52.559
think it might also just change a little bit

00:11:52.559 --> 00:11:55.240
how people try and measure this i mean you've

00:11:55.240 --> 00:11:59.259
got hrefs and um semrush and all these other

00:11:59.259 --> 00:12:02.840
platforms trying to offer share of ai voice tracking

00:12:02.840 --> 00:12:07.429
which I suspect it's not that useful right now

00:12:07.429 --> 00:12:09.629
because of just how dynamic these systems are

00:12:09.629 --> 00:12:11.870
even compared to search. But I think people are

00:12:11.870 --> 00:12:14.169
probably going to still be desperate to figure

00:12:14.169 --> 00:12:16.889
out if their brands are being mentioned in AI

00:12:16.889 --> 00:12:20.250
answers. So maybe we see some platform development

00:12:20.250 --> 00:12:23.789
there. Maybe we see, you know, ChatGPT offering

00:12:23.789 --> 00:12:26.509
the equivalent of a Google search console so

00:12:26.509 --> 00:12:28.250
that we can actually better understand how our

00:12:28.250 --> 00:12:31.259
brands are being. are being mentioned when people

00:12:31.259 --> 00:12:33.679
are searching for you know problems and solutions

00:12:33.679 --> 00:12:37.840
related to what we do um and then i think we'll

00:12:37.840 --> 00:12:39.639
continue to see people lean into high quality

00:12:39.639 --> 00:12:43.519
content to try and make sure that theirs is the

00:12:43.519 --> 00:12:46.759
brand that gets cited especially you know in

00:12:46.759 --> 00:12:50.019
a world of low quality ai filler and i think

00:12:50.019 --> 00:12:52.659
it will be things like original research proprietary

00:12:52.659 --> 00:12:54.700
information that you can't really get elsewhere

00:12:54.700 --> 00:12:57.679
i think humans find that valuable and i think

00:12:58.029 --> 00:13:00.289
ai chatbots will realize that humans find that

00:13:00.289 --> 00:13:02.830
valuable and that's what they'll look for to

00:13:02.830 --> 00:13:05.549
surface a lot of the educational type content

00:13:05.549 --> 00:13:09.330
that they provide in their answers so what do

00:13:09.330 --> 00:13:11.649
you reckon martin on the uh continued growth

00:13:11.649 --> 00:13:14.490
of interest in geo let's call it for simplicity

00:13:14.490 --> 00:13:17.129
well it's not going away that's for sure every

00:13:17.129 --> 00:13:20.149
workshop that i have has people saying oh how

00:13:20.149 --> 00:13:22.710
do i need to optimize for visibility in this

00:13:22.710 --> 00:13:25.450
but i guess following on from that then paul

00:13:25.450 --> 00:13:30.870
do you think the seo role will just expand to

00:13:30.870 --> 00:13:33.129
encompass this or do you think it's going to

00:13:33.129 --> 00:13:36.870
be a new role a new specialism in and of itself

00:13:36.870 --> 00:13:40.629
yeah that's a great question i think the simple

00:13:40.629 --> 00:13:44.070
answer is it's the same role with some additional

00:13:44.070 --> 00:13:48.009
strings required to add to the bow do you know

00:13:48.009 --> 00:13:51.110
i actually think it potentially makes life exciting

00:13:51.110 --> 00:13:53.450
for seos again maybe they've always considered

00:13:53.450 --> 00:13:55.909
their life a bit challenging and exciting but

00:13:56.899 --> 00:14:01.240
I think they're already very used to trying to

00:14:01.240 --> 00:14:03.379
reverse engineer and figure out what we needed

00:14:03.379 --> 00:14:06.480
to do to be top of Google search. And now they're

00:14:06.480 --> 00:14:08.320
having to do that to make sure their brands get

00:14:08.320 --> 00:14:12.840
cited in AI answers. And I suspect that a lot

00:14:12.840 --> 00:14:15.419
of the tactics and tools they've already used

00:14:15.419 --> 00:14:18.980
to using still apply. And I also expect there

00:14:18.980 --> 00:14:22.799
to be the ebb and flow of the currents, right?

00:14:22.879 --> 00:14:24.580
There's going to be an opportunity. We've talked

00:14:24.580 --> 00:14:26.580
about this on the pod before. If we get loads

00:14:26.580 --> 00:14:29.820
of product reviews and ChatGPT is obsessed with

00:14:29.820 --> 00:14:31.559
product reviews and it thinks the best products

00:14:31.559 --> 00:14:33.159
are definitely the ones that have the most reviews

00:14:33.159 --> 00:14:35.299
and the highest average review. And that's the

00:14:35.299 --> 00:14:37.539
one that it recommends every time. People obsess

00:14:37.539 --> 00:14:41.120
about reviews and then ChatGPT and OpenAI will

00:14:41.120 --> 00:14:42.940
be like, oh, we need to dial that down a bit

00:14:42.940 --> 00:14:44.879
then because that's been gained. And I expect

00:14:44.879 --> 00:14:48.100
to see all of that play out. And who better to

00:14:48.100 --> 00:14:50.720
face that challenge than people who've been doing

00:14:50.720 --> 00:14:52.419
it in a slightly different form for the last

00:14:52.419 --> 00:14:54.620
decade? I don't know. What do you think? Well.

00:14:55.569 --> 00:14:57.990
forget what i think i'm really interested to

00:14:57.990 --> 00:15:01.110
know how we measure this in 12 months time what's

00:15:01.110 --> 00:15:03.690
the metric what's the what's the done not done

00:15:03.690 --> 00:15:06.769
in eos terms paul like whether or not it came

00:15:06.769 --> 00:15:08.590
true or not yeah i think that's a really good

00:15:08.590 --> 00:15:11.870
question so i think what we should expect to

00:15:11.870 --> 00:15:15.649
see is an increase in the amount of traffic that

00:15:15.649 --> 00:15:17.629
we're getting from ai chatbots so if you're not

00:15:17.629 --> 00:15:19.809
tracking this already it's fairly easy to do

00:15:19.809 --> 00:15:21.590
you should be able to see referral traffic from

00:15:21.590 --> 00:15:25.190
example um ChatGPT, you should be able to see

00:15:25.190 --> 00:15:27.889
referral traffic from Gemini, Perplexity. A lot

00:15:27.889 --> 00:15:30.850
of the brands that I have access to, I can see

00:15:30.850 --> 00:15:32.529
that there's a small amount of that traffic and

00:15:32.529 --> 00:15:35.009
it's going up. And interestingly, so Biostrata,

00:15:35.049 --> 00:15:37.730
the agency I co -founded, we currently get 8

00:15:37.730 --> 00:15:42.110
% of our traffic from AI chatbots, which is actually

00:15:42.110 --> 00:15:44.409
pretty huge, I think, compared to a lot of other

00:15:44.409 --> 00:15:48.580
brands. Most of the brands, Hayler on the podcast

00:15:48.580 --> 00:15:50.899
I was on yesterday, we were talking about, she

00:15:50.899 --> 00:15:52.740
said most of the brands she's looked at recently

00:15:52.740 --> 00:15:55.299
is more like 1%. So it's still very, very small.

00:15:55.399 --> 00:15:57.740
But we should see that grow if this is true.

00:15:58.940 --> 00:16:00.620
Assuming that there's a mechanism that makes

00:16:00.620 --> 00:16:03.019
it easier for you to click through, which we

00:16:03.019 --> 00:16:06.000
don't know for sure. And I do think people are

00:16:06.000 --> 00:16:08.460
going to try and find a way to measure some sort

00:16:08.460 --> 00:16:12.279
of AI share of voice. I really don't believe

00:16:12.279 --> 00:16:14.820
the platform is there to do it yet. we would

00:16:14.820 --> 00:16:17.620
need to see a much better platform released,

00:16:17.700 --> 00:16:20.379
in my opinion. But I think people are going to

00:16:20.379 --> 00:16:23.279
want to try and measure it. It will end up being

00:16:23.279 --> 00:16:27.820
the equivalent of trying to track your keyword

00:16:27.820 --> 00:16:30.440
rankings, right? And trying to see how often

00:16:30.440 --> 00:16:33.620
you're mentioned in relevant queries. But I think

00:16:33.620 --> 00:16:35.419
the gold standard is going to be more referral

00:16:35.419 --> 00:16:37.960
traffic, at least to begin with. We'll talk about

00:16:37.960 --> 00:16:39.639
that as part of a different prediction in a moment.

00:16:41.059 --> 00:16:43.220
And tracking the conversion rate of that traffic,

00:16:43.340 --> 00:16:45.379
like do those people actually go and buy something

00:16:45.379 --> 00:16:48.139
from you? Good old marketing fundamentals, right?

00:16:48.820 --> 00:16:52.320
Are we getting good volume and is it converting

00:16:52.320 --> 00:16:56.440
into something of value? Yeah, it's like you

00:16:56.440 --> 00:16:58.779
said, so much of this stuff is going to be about

00:16:58.779 --> 00:17:00.279
applying fundamentals. I think where it will

00:17:00.279 --> 00:17:03.960
get actually crazy is when things become more

00:17:03.960 --> 00:17:07.359
agentic and you type something in chat GPT. And

00:17:07.359 --> 00:17:09.400
it doesn't even show you the product recommendations.

00:17:09.640 --> 00:17:11.720
It finds the three products that it thinks are

00:17:11.720 --> 00:17:14.619
best. It engages in a price negotiation of the

00:17:14.619 --> 00:17:17.559
agent of one of the merchants. And then it buys

00:17:17.559 --> 00:17:19.819
it for you. And then it turns up at your house.

00:17:20.400 --> 00:17:23.880
As a marketer, how are you going to measure that?

00:17:23.960 --> 00:17:26.680
Is that traffic? Is that conversions? How do

00:17:26.680 --> 00:17:29.839
you measure an AI agent conversion versus a human

00:17:29.839 --> 00:17:32.599
conversion? So I think all of these things hold

00:17:32.599 --> 00:17:34.579
true for now. Well, I think they'll probably

00:17:34.579 --> 00:17:36.930
hold. true until the end of the year asterisk

00:17:36.930 --> 00:17:42.230
we'll speak later um but yes focus on the fundamentals

00:17:42.230 --> 00:17:46.369
for now maybe the world gets a bit shaken over

00:17:46.369 --> 00:17:49.269
the next 24 months oh what have you got for us

00:17:49.269 --> 00:17:52.430
on your prediction number two mine so mine is

00:17:52.430 --> 00:17:55.329
moving into the world of model development and

00:17:55.329 --> 00:17:59.910
my prediction is that there will be no new flagship

00:17:59.910 --> 00:18:04.799
number jumps so to speak from any of the major

00:18:04.799 --> 00:18:07.900
model labs. So I don't think we're going to see

00:18:07.900 --> 00:18:13.519
OpenAI launch GPT -6. I don't think we're going

00:18:13.519 --> 00:18:17.279
to see Gemini release Gemini 4, or should I say

00:18:17.279 --> 00:18:20.599
Google release Gemini 4. And I don't think we're

00:18:20.599 --> 00:18:26.839
going to see Lord release Opus 5. However, I

00:18:26.839 --> 00:18:32.529
do think we'll see Lord Haiku 5. Will we get

00:18:32.529 --> 00:18:35.190
a new Grok? I'm not paying attention to that,

00:18:35.250 --> 00:18:40.369
to be perfectly honest. We might hear that there

00:18:40.369 --> 00:18:42.890
is a new Grok. Elon will tell us that there is

00:18:42.890 --> 00:18:45.730
one coming in the next six days that's going

00:18:45.730 --> 00:18:48.809
to clean your car and fly you to the moon. It

00:18:48.809 --> 00:18:52.930
may or may not happen. I choose, due to my own

00:18:52.930 --> 00:18:58.289
personal set of values and principles, to ignore

00:18:58.289 --> 00:19:01.920
that side of the AI landscape. for the time being

00:19:01.920 --> 00:19:04.839
this really hamstring your ability to create

00:19:04.839 --> 00:19:07.160
inappropriate nude pictures then i'm afraid martin

00:19:07.160 --> 00:19:10.380
but um that's probably for the best what what

00:19:10.380 --> 00:19:12.160
makes you think this do you think we're just

00:19:12.160 --> 00:19:15.700
outside of the cycle or yeah it's as simple as

00:19:15.700 --> 00:19:17.359
that to be perfectly honest if you look at how

00:19:17.359 --> 00:19:20.460
long it took for them to go from four to five

00:19:20.460 --> 00:19:24.740
i think we'll see the the point gains um progressing

00:19:24.740 --> 00:19:30.119
so i think we'll see 5 .2 maybe turn into 5 .5

00:19:30.920 --> 00:19:36.240
I think we'll see a Gemini turn into a Gemini

00:19:36.240 --> 00:19:42.099
3 .5 at some point in the year. If I was really

00:19:42.099 --> 00:19:46.619
pressed, I could see Google launching Gemini

00:19:46.619 --> 00:19:52.640
for flash maybe. Um, and doing one of the smaller

00:19:52.640 --> 00:19:58.079
models in a number leap. But I just don't see

00:19:58.079 --> 00:20:00.420
it. I think the cycle will tip over. They'll

00:20:00.420 --> 00:20:05.460
give each of these 18 months to run. And that's

00:20:05.460 --> 00:20:07.680
partly because I don't think we've got close

00:20:07.680 --> 00:20:12.759
to saturating the use cases for where they're

00:20:12.759 --> 00:20:15.480
currently at. I think that there's still quite

00:20:15.480 --> 00:20:18.160
a lot of runway for them. And I know that all

00:20:18.160 --> 00:20:24.799
of the model providers, they talk about reserving

00:20:24.799 --> 00:20:29.079
the big... leap for some new capability and at

00:20:29.079 --> 00:20:32.400
the moment i think the agentic piece the computer

00:20:32.400 --> 00:20:36.779
use metric is the one that is really interesting

00:20:36.779 --> 00:20:41.059
across the board like all the the leap in gpt5

00:20:41.059 --> 00:20:44.839
the leap in flawed and all of them have seen

00:20:44.839 --> 00:20:47.579
big leaps in computer use and that hasn't really

00:20:47.579 --> 00:20:51.779
rolled out into the world yet so i think once

00:20:51.779 --> 00:20:54.990
that actual application of it and people are

00:20:54.990 --> 00:20:57.569
going oh my god this this new skill that it's

00:20:57.569 --> 00:21:01.609
got is actually pretty good um and then people

00:21:01.609 --> 00:21:03.609
start going when's the next one coming we need

00:21:03.609 --> 00:21:06.369
more of this because you know the general knowledge

00:21:06.369 --> 00:21:09.710
and intelligence of these things is already typically

00:21:09.710 --> 00:21:15.529
exceeding what i would say 90 of use cases 90

00:21:15.529 --> 00:21:20.109
plus use cases really needs so yeah i think once

00:21:20.109 --> 00:21:25.359
the the computer use that element is being more

00:21:25.359 --> 00:21:29.579
rolled out and more seen and felt widely by consumers

00:21:29.579 --> 00:21:32.700
and enterprise, then they'll say, right, now

00:21:32.700 --> 00:21:35.619
it's time for us to take the next numerical leap.

00:21:35.819 --> 00:21:37.759
I think I agree with you, but in the interest

00:21:37.759 --> 00:21:41.579
of making this a dynamic conversation, the only

00:21:41.579 --> 00:21:43.599
reason I can think, there's a couple of things

00:21:43.599 --> 00:21:45.859
that spring to mind. The first one is some of

00:21:45.859 --> 00:21:50.339
the bigger GPU clusters, when they start to come

00:21:50.339 --> 00:21:54.049
online, I think maybe I shouldn't mention this

00:21:54.049 --> 00:21:55.410
now because I don't want to trigger you again,

00:21:55.529 --> 00:21:58.230
but I think Elon's got one coming online in the

00:21:58.230 --> 00:22:00.890
next like month or two, hasn't he? I'm sure he

00:22:00.890 --> 00:22:04.789
has. I'm sure he's told you that a week ago and

00:22:04.789 --> 00:22:07.190
it's just like self -driving cars. It's just

00:22:07.190 --> 00:22:09.250
around the corner. He's going to be living on

00:22:09.250 --> 00:22:12.750
Mars in 2026. I'm sure he said that in 2015.

00:22:13.470 --> 00:22:17.349
Just need to dial it back. Yeah, chill. And be,

00:22:17.450 --> 00:22:21.079
you know, be calm. Yeah, there'll be gigawatt

00:22:21.079 --> 00:22:25.319
clusters coming online, loads of GPUs, incrementally

00:22:25.319 --> 00:22:27.240
every training run, assuming they're still even

00:22:27.240 --> 00:22:29.579
bothering to try and train new models, because

00:22:29.579 --> 00:22:31.279
I think they've sucked up a lot of the high quality

00:22:31.279 --> 00:22:33.279
data that's available, if I understand correctly.

00:22:33.420 --> 00:22:35.859
But let's say they're still doing big, ever bigger

00:22:35.859 --> 00:22:38.740
training runs. One assumes in theory, maybe they

00:22:38.740 --> 00:22:44.380
could go faster if they had 10x, 100x more, I'm

00:22:44.380 --> 00:22:46.299
going to call it GPU power, because of course,

00:22:46.299 --> 00:22:49.119
we're also getting better GPUs, not just more.

00:22:49.789 --> 00:22:52.609
GPUs. So in theory, that might allow a training

00:22:52.609 --> 00:22:55.490
run to get done before the end of the year. I

00:22:55.490 --> 00:22:57.470
still think you're probably right. The other

00:22:57.470 --> 00:23:01.230
thing is your point about giving models new numbers

00:23:01.230 --> 00:23:04.170
when they do something cool they can't do now.

00:23:04.849 --> 00:23:08.450
And we already saw some of this with what should

00:23:08.450 --> 00:23:11.630
have probably been GPT -5 wasn't. It was GPT

00:23:11.630 --> 00:23:14.289
-4 .5 probably because its capabilities were

00:23:14.289 --> 00:23:17.740
not quite what was hoped for. And then OpenAI

00:23:17.740 --> 00:23:21.460
always also had that weird naming convention

00:23:21.460 --> 00:23:24.559
when thinking came out, right, where we had the

00:23:24.559 --> 00:23:28.420
O series of models. So maybe there's a research

00:23:28.420 --> 00:23:31.259
breakthrough and we get a new capability or,

00:23:31.319 --> 00:23:34.519
you know, agents become much better and we get

00:23:34.519 --> 00:23:39.640
GPT -A1 or something, which is something completely

00:23:39.640 --> 00:23:42.759
different that does other cool stuff. Without

00:23:42.759 --> 00:23:44.940
being able to see inside the machine, I think

00:23:44.940 --> 00:23:46.900
that's quite hard to predict. I think probably

00:23:46.900 --> 00:23:51.039
your prediction's safe. What does this mean for

00:23:51.039 --> 00:23:52.440
marketers? What do we have to keep an eye on

00:23:52.440 --> 00:23:57.740
here? I think it means that you can, for the

00:23:57.740 --> 00:24:00.759
time being, and this would be my advice, regardless,

00:24:01.079 --> 00:24:04.160
really focus on squeezing as much value out of

00:24:04.160 --> 00:24:06.079
the tools that are in front of you today. The

00:24:06.079 --> 00:24:10.619
capabilities of these models is... amazing, the

00:24:10.619 --> 00:24:13.220
agentic capabilities in particular. But if you're

00:24:13.220 --> 00:24:17.779
still playing around with these models for content

00:24:17.779 --> 00:24:23.059
creation and maybe helping draft awkward emails

00:24:23.059 --> 00:24:27.220
or ideation, you are barely scratching the surface

00:24:27.220 --> 00:24:34.680
of what these tools are capable of. Lord. on

00:24:34.680 --> 00:24:38.200
its own has an amazing set of skills now for

00:24:38.200 --> 00:24:43.680
document creation. In fact, GPT 5 .2 has a similar

00:24:43.680 --> 00:24:46.680
thing as well. For the first time this week,

00:24:46.900 --> 00:24:51.140
I asked it to create a PowerPoint for me, something

00:24:51.140 --> 00:24:57.339
that I've been doing with GPT, well, since it's

00:24:57.339 --> 00:25:01.460
had Code Interpreter, really. And it's been able

00:25:01.460 --> 00:25:05.279
to create quite basic looking powerpoint files

00:25:05.279 --> 00:25:08.000
they were very blank and it had used the basic

00:25:08.000 --> 00:25:11.039
template yesterday in a workshop i asked it to

00:25:11.039 --> 00:25:14.119
create a powerpoint file after 10 minutes of

00:25:14.119 --> 00:25:16.799
watching it in thinking mode still going i said

00:25:16.799 --> 00:25:18.819
oh don't worry we'll come back to that later

00:25:18.819 --> 00:25:22.339
on we broke for lunch when we came back it had

00:25:22.339 --> 00:25:24.700
finished thinking for 50 minutes and it created

00:25:24.700 --> 00:25:29.140
a powerpoint that was fully designed with graphics

00:25:30.990 --> 00:25:34.289
with pngs embedded within it created its own

00:25:34.289 --> 00:25:37.970
layouts as a powerpoint file it was way beyond

00:25:37.970 --> 00:25:43.569
what i had seen it do before now i can easily

00:25:43.569 --> 00:25:45.950
imagine a world where if i had known it could

00:25:45.950 --> 00:25:49.410
do that and i had specified things like the look

00:25:49.410 --> 00:25:52.049
and feel the style guide of the powerpoint template

00:25:52.049 --> 00:25:55.650
it would have done it to the to the style that

00:25:55.650 --> 00:25:59.240
i instructed But that's where we're going. These

00:25:59.240 --> 00:26:02.380
can actually execute tasks on, again, that was

00:26:02.380 --> 00:26:06.200
a 50 minute thinking exercise. It was, it's definitely

00:26:06.200 --> 00:26:08.440
the longest phase of thinking that I've seen

00:26:08.440 --> 00:26:11.839
ChatGPT do for me. That's so cool. Cause that's

00:26:11.839 --> 00:26:16.039
like quasi agentic, right? It's done a load of

00:26:16.039 --> 00:26:18.920
stuff for you, but it's not what we're currently

00:26:18.920 --> 00:26:23.200
thinking is agentic in terms of tool use, like

00:26:23.200 --> 00:26:28.680
almost like. overt tool use mcp browser agents

00:26:28.680 --> 00:26:30.920
that click on things for you it's like it did

00:26:30.920 --> 00:26:34.339
a load of extra work for you and gave you this

00:26:34.339 --> 00:26:38.160
amazing output that's super interesting okay

00:26:38.160 --> 00:26:41.240
i agree there's loads of power there to uh to

00:26:41.240 --> 00:26:44.940
tap into before we get more model numbers um

00:26:44.940 --> 00:26:49.180
all right next trend so my next prediction not

00:26:49.180 --> 00:26:54.720
trend prediction is by the end of 2026 You'll

00:26:54.720 --> 00:26:59.380
be executing purchases inside chatbot. So you'll

00:26:59.380 --> 00:27:02.359
search for, give me some ideas for a product

00:27:02.359 --> 00:27:04.460
for this. It'll make some recommendations. You'll

00:27:04.460 --> 00:27:06.759
click on the one that you want to buy and it

00:27:06.759 --> 00:27:08.880
will just be purchased. You won't go to a website.

00:27:08.980 --> 00:27:12.180
You won't fill in checkout information. It will

00:27:12.180 --> 00:27:15.779
just buy it for you. And it's interesting because

00:27:15.779 --> 00:27:18.859
this is a prediction show, not a new show. But

00:27:18.859 --> 00:27:22.559
this week, Google did actually launch. a new

00:27:22.559 --> 00:27:29.240
open sort of open source set of um tools and

00:27:29.240 --> 00:27:32.240
and protocols for this type of experience that

00:27:32.240 --> 00:27:35.680
they call the agent payments protocol that they

00:27:35.680 --> 00:27:40.240
built with like shopify etsy wayfair target walmart

00:27:40.240 --> 00:27:44.799
um and you know american express and mastercard

00:27:44.799 --> 00:27:46.839
and stripe all these people as you know all of

00:27:46.839 --> 00:27:50.779
these important merchants and payment gateways

00:27:50.779 --> 00:27:52.960
have all said they're going to support it so

00:27:52.960 --> 00:27:55.519
i do think by the end of the year that we'll

00:27:55.519 --> 00:27:59.640
have that and oh go on no go on mine what's your

00:27:59.640 --> 00:28:00.940
thought i was going to say i think this one's

00:28:00.940 --> 00:28:04.500
a slam dunk in so much as google have made that

00:28:04.500 --> 00:28:08.039
announcement and on the 8th of january co -pilot

00:28:08.039 --> 00:28:11.180
checkout and brand agents was announced by microsoft

00:28:11.180 --> 00:28:14.920
as well in partnership with paypal shopify and

00:28:14.920 --> 00:28:19.740
stripe And yeah, I, I think this is going to

00:28:19.740 --> 00:28:23.700
become the de facto standard for online shopping

00:28:23.700 --> 00:28:30.319
experiences. I would also say that the, the brand

00:28:30.319 --> 00:28:35.980
agent element in particular is an interesting

00:28:35.980 --> 00:28:40.740
tool to watch out for if you're a marketer, because

00:28:40.740 --> 00:28:44.819
they are launching within the. So Microsoft Clarity,

00:28:44.920 --> 00:28:48.039
the analytics and user experience tool that Microsoft

00:28:48.039 --> 00:28:51.839
has given away for free for a long time, it's

00:28:51.839 --> 00:28:54.599
a kind of competitor to Hotjar. It's that kind

00:28:54.599 --> 00:28:57.119
of a tool for anyone that hasn't used it before.

00:28:57.940 --> 00:29:02.380
Within the Clarity brand experience and product

00:29:02.380 --> 00:29:04.859
experience, they are launching this brand agent,

00:29:05.160 --> 00:29:09.759
which is basically a way of integrating a co

00:29:09.759 --> 00:29:12.240
-pilot assistant that understands your product

00:29:12.240 --> 00:29:15.910
range. and can have an interactive chat with

00:29:15.910 --> 00:29:18.309
somebody on your website and integrates with

00:29:18.309 --> 00:29:22.430
the likes of Shopify. So you can just integrate

00:29:22.430 --> 00:29:27.849
seamlessly your product stream and then have

00:29:27.849 --> 00:29:31.789
an AI assistant guiding customers towards the

00:29:31.789 --> 00:29:33.289
checkout. And that was something that they announced

00:29:33.289 --> 00:29:36.269
last week. I think you highlight a load of really

00:29:36.269 --> 00:29:38.509
important things there that marketers are going

00:29:38.509 --> 00:29:40.529
to need to be. thinking about in order to be

00:29:40.529 --> 00:29:42.509
able to be ready for this shift maybe less so

00:29:42.509 --> 00:29:44.569
in the b2b world that you and i tend to operate

00:29:44.569 --> 00:29:48.609
in but certainly in b2c i think it's i think

00:29:48.609 --> 00:29:51.329
it's going to have ramifications i'm going to

00:29:51.329 --> 00:29:53.970
extend my prediction even further then i'm going

00:29:53.970 --> 00:29:58.650
to say there is going to be purchases made where

00:29:58.650 --> 00:30:01.369
people don't review the products i'll give you

00:30:01.369 --> 00:30:05.190
an example i am going to say this is not a great

00:30:05.190 --> 00:30:06.930
example but it's the one that instantly comes

00:30:06.930 --> 00:30:11.410
to mind ChatGPT, buy Martin a product for his

00:30:11.410 --> 00:30:15.789
birthday and send it. And I'm not going to look

00:30:15.789 --> 00:30:18.049
at the products, not going to enter any data.

00:30:18.650 --> 00:30:21.170
It's going to buy it and it'll arrive at your

00:30:21.170 --> 00:30:24.690
door and my credit card will be charged for it.

00:30:25.029 --> 00:30:27.069
Because I think we're going to start to enter

00:30:27.069 --> 00:30:32.789
the world of agentic purchases where we identify

00:30:32.789 --> 00:30:35.849
the broad outcome for certain. requirements and

00:30:35.849 --> 00:30:38.609
product types and we don't even approve the purchase

00:30:38.609 --> 00:30:42.789
and there's going to be a bunch of stuff that

00:30:42.789 --> 00:30:44.990
needs to go with that right like i'm thinking

00:30:44.990 --> 00:30:48.970
yeah you'll have to probably approve a bunch

00:30:48.970 --> 00:30:50.950
of purchases for a few months just so the system

00:30:50.950 --> 00:30:54.690
can learn how often it gets your preferences

00:30:54.690 --> 00:30:56.890
when it you know this is the product i would

00:30:56.890 --> 00:30:58.309
have bought is this the product you would have

00:30:58.309 --> 00:31:00.150
bought and you have to tell it yes or no so we

00:31:00.150 --> 00:31:03.440
can learn you know a bit more about you um And

00:31:03.440 --> 00:31:05.539
there'll be other things like budget caps and

00:31:05.539 --> 00:31:07.380
stuff like, no, I can't just go and buy something

00:31:07.380 --> 00:31:11.319
for you that's like, you know, $5 ,000. You know,

00:31:11.319 --> 00:31:13.180
to begin with, I think the way that we'll all

00:31:13.180 --> 00:31:15.619
get comfortable with this is, oh, my toothpaste

00:31:15.619 --> 00:31:18.180
is running out. Go get me some more toothpaste.

00:31:18.579 --> 00:31:20.259
And maybe it knows what toothpaste you've had

00:31:20.259 --> 00:31:21.740
in the past. And so it just goes, buys it for

00:31:21.740 --> 00:31:24.440
you. Or maybe it finds out that there's a better

00:31:24.440 --> 00:31:26.200
toothpaste that's currently on sale. And then

00:31:26.200 --> 00:31:28.079
it buys that for you instead, because it knows

00:31:28.079 --> 00:31:30.339
that cavity protection is the most important

00:31:30.339 --> 00:31:33.440
thing for you based on. The last bit of toothpaste

00:31:33.440 --> 00:31:35.240
that you bought. And I think we're going to start

00:31:35.240 --> 00:31:40.319
to see a bunch of this stuff emerge. So how will

00:31:40.319 --> 00:31:44.039
we know if it's true? I think you'll be able

00:31:44.039 --> 00:31:46.319
to click. Firstly, you'll be able to click buy

00:31:46.319 --> 00:31:50.759
now in an AI chatbot experience like ChatGPT

00:31:50.759 --> 00:31:55.059
or Gemini. And it will just say, bing, done.

00:31:56.039 --> 00:32:00.420
And if the second prediction is true, we'll see.

00:32:01.239 --> 00:32:04.200
it select the product and you say go buy me something

00:32:04.200 --> 00:32:06.440
like this and then it just says bing done and

00:32:06.440 --> 00:32:08.700
it's not you didn't even select the product you

00:32:08.700 --> 00:32:11.400
just gave it like an overall brief what do you

00:32:11.400 --> 00:32:12.880
think about that second part martin do you reckon

00:32:12.880 --> 00:32:15.180
that's gonna happen yeah i like that and i like

00:32:15.180 --> 00:32:17.900
if i think about that from a consumer perspective

00:32:17.900 --> 00:32:21.720
i can see the way that platforms like the payment

00:32:21.720 --> 00:32:25.819
gateway stripe paypal they are probably going

00:32:25.819 --> 00:32:28.960
to end up holding a lot of the the preference

00:32:28.960 --> 00:32:34.200
data I can see them storing purchase profiles

00:32:34.200 --> 00:32:39.980
of you as a consumer that they then basically

00:32:39.980 --> 00:32:46.160
plug into the agents on different commerce platforms.

00:32:46.380 --> 00:32:49.819
And interestingly, and I did wonder what was

00:32:49.819 --> 00:32:52.599
going on there. This week when I went into PayPal,

00:32:52.799 --> 00:32:59.019
they offered me a PayPal debit card, which is

00:32:59.019 --> 00:33:05.730
linked. to my main bank account. So it's just

00:33:05.730 --> 00:33:09.230
going to debit my ordinary account, but there's

00:33:09.230 --> 00:33:13.490
a kind of cash back. I think it's 1 % cash back

00:33:13.490 --> 00:33:16.269
incentive to it. And I thought, Oh, that's interesting.

00:33:16.390 --> 00:33:19.250
I wonder what the play is here. And it's making

00:33:19.250 --> 00:33:22.029
me think that, Hmm, that's a great way for them

00:33:22.029 --> 00:33:26.750
to harvest data, um, to get you to spend all

00:33:26.750 --> 00:33:30.569
of your money through. paypal rather than just

00:33:30.569 --> 00:33:34.029
the online transactions at certain uh certain

00:33:34.029 --> 00:33:37.170
retailers but every time you go to a store in

00:33:37.170 --> 00:33:41.849
person you go to the cinema every time you tap

00:33:41.849 --> 00:33:44.750
and pay on the underground or anything like that

00:33:44.750 --> 00:33:47.390
they can build that profile of you yeah i mean

00:33:47.390 --> 00:33:49.170
there's a load of second order effects to this

00:33:49.170 --> 00:33:52.269
right because it hasn't happened this way because

00:33:52.269 --> 00:33:54.309
we can see this news coming out this week already

00:33:54.309 --> 00:33:58.150
but imagine that this agentic payment payment

00:33:58.150 --> 00:34:02.809
framework emerged and it was being led by paypal

00:34:02.809 --> 00:34:06.529
and no one else offered it for you know six months

00:34:06.529 --> 00:34:10.349
and just how much paypal could then come to dominate

00:34:10.349 --> 00:34:14.369
the payment gateway market especially for maybe

00:34:14.369 --> 00:34:17.110
tech savvy consumers who want to be able to pay

00:34:17.110 --> 00:34:19.650
like this and it doesn't surprise me that hasn't

00:34:19.650 --> 00:34:21.250
happened because everyone's like whoa we cannot

00:34:21.250 --> 00:34:25.050
afford to be missing that train I think it also

00:34:25.050 --> 00:34:27.090
brings a bunch of other second order consequences

00:34:27.090 --> 00:34:31.869
in line with what you said, because even if we

00:34:31.869 --> 00:34:35.230
look at supermarkets and their loyalty cards,

00:34:35.369 --> 00:34:40.969
right, tracking what people buy is not new. But

00:34:40.969 --> 00:34:43.909
in order for this system that I've predicted

00:34:43.909 --> 00:34:48.730
to work, the technology is going to need an ever

00:34:48.730 --> 00:34:50.849
more granular understanding of our buying habits.

00:34:51.510 --> 00:34:54.940
And to be honest. the richest understanding they

00:34:54.940 --> 00:34:58.639
can get of us the better because the better they're

00:34:58.639 --> 00:35:00.159
going to operate the more seamless it's going

00:35:00.159 --> 00:35:02.059
to be we're not going to trust a system that's

00:35:02.059 --> 00:35:06.260
too error -strewn right so having that data becomes

00:35:06.260 --> 00:35:09.480
critical and everybody's going to do whatever

00:35:09.480 --> 00:35:12.199
they can to get as much of it as possible and

00:35:12.199 --> 00:35:16.039
this is paypal's play my guess would still be

00:35:17.480 --> 00:35:21.019
That your tool of choice, like the pursuit of

00:35:21.019 --> 00:35:24.420
ChatGPT and Gemini to be your assistant of record

00:35:24.420 --> 00:35:28.159
is critical. Because that's where they can build

00:35:28.159 --> 00:35:30.239
up this massive picture of you. I think the assistant

00:35:30.239 --> 00:35:32.380
is going to know a lot about me and a lot about

00:35:32.380 --> 00:35:35.679
you and a lot about hundreds of millions, if

00:35:35.679 --> 00:35:38.039
not billions of people. So they can do all of

00:35:38.039 --> 00:35:41.760
the aggregate. predictions that probably ai will

00:35:41.760 --> 00:35:43.500
be quite good at sorting through all of that

00:35:43.500 --> 00:35:45.500
data to try and figure out you know i'll start

00:35:45.500 --> 00:35:48.079
seeing things turn up at my house that martin

00:35:48.079 --> 00:35:50.659
buys because it thinks that i'm just like martin

00:35:50.659 --> 00:35:52.719
and so i probably want that thing as well for

00:35:52.719 --> 00:35:54.699
those of you that know martin that is an extremely

00:35:54.699 --> 00:35:57.840
scary thought for me you want to make sure you

00:35:57.840 --> 00:36:01.860
open those packages away from anyone else in

00:36:01.860 --> 00:36:04.400
the household do you know it would just be all

00:36:04.400 --> 00:36:08.079
of the uh tech that you buy on a kickstarter

00:36:08.079 --> 00:36:11.019
that maybe becomes massive and maybe it'll be

00:36:11.019 --> 00:36:12.780
another one that meta buys and you'll just get

00:36:12.780 --> 00:36:15.679
your money back oh that is an important point

00:36:15.679 --> 00:36:18.260
and an important clarification in the last episode

00:36:18.260 --> 00:36:21.860
i did say that after limitless were acquired

00:36:21.860 --> 00:36:27.599
by meta i wouldn't get a refund from my investment

00:36:27.599 --> 00:36:31.099
i'm going to call it an investment in the pendant

00:36:31.099 --> 00:36:35.139
ai but they did they refunded me in full so hats

00:36:35.139 --> 00:36:38.239
off to them And I take it all back. I love Meta.

00:36:38.260 --> 00:36:41.139
They're a very responsible consumer -first company.

00:36:41.480 --> 00:36:44.820
Oh, ye of little faith. So what can marketers

00:36:44.820 --> 00:36:48.639
do? If you work in B2C, you should be starting

00:36:48.639 --> 00:36:53.239
to borderline obsess about this now because I

00:36:53.239 --> 00:36:55.800
do think it's coming probably quite quickly.

00:36:55.860 --> 00:37:00.260
Once that payment capability is built in, I can

00:37:00.260 --> 00:37:03.619
see certainly certain age demographics, I can

00:37:03.619 --> 00:37:06.940
see their behavior change quite quickly. potentially

00:37:06.940 --> 00:37:11.860
um so yeah i think probably making sure that

00:37:11.860 --> 00:37:16.019
your i'm going to call it your website but i'm

00:37:16.019 --> 00:37:17.480
sure there'll be a time when it won't be your

00:37:17.480 --> 00:37:19.840
website it'll probably be handled by apis that

00:37:19.840 --> 00:37:21.719
maybe know your stock levels and stuff like that

00:37:21.719 --> 00:37:25.139
but just making sure that you have all of your

00:37:25.139 --> 00:37:28.860
product information machine readable easy for

00:37:29.340 --> 00:37:32.239
AI chatbots to understand what would be the appropriate

00:37:32.239 --> 00:37:35.119
product for an appropriate requirement. Certainly

00:37:35.119 --> 00:37:39.539
the days of really unhelpful product names or

00:37:39.539 --> 00:37:43.039
names and identities based on number strings

00:37:43.039 --> 00:37:45.400
that don't mean anything. Like I can see that

00:37:45.400 --> 00:37:48.679
would be a problem if that's how you still manage

00:37:48.679 --> 00:37:51.699
your product inventory, for example. And then

00:37:51.699 --> 00:37:56.239
I think just starting to try and measure some

00:37:56.239 --> 00:37:58.070
of that. Again, where's that going to be done?

00:37:58.230 --> 00:38:00.429
I think you'll probably be able to see the number

00:38:00.429 --> 00:38:04.730
of purchases in your, let's call it ChatGPT search

00:38:04.730 --> 00:38:08.050
console or whatever it may be. Yeah, ChatGPT

00:38:08.050 --> 00:38:11.469
Merchant Center. Right, right. Yeah. I would

00:38:11.469 --> 00:38:13.590
imagine we're probably going to need to see that

00:38:13.590 --> 00:38:17.389
emerge actually as a platform and a tool because...

00:38:19.159 --> 00:38:21.019
People are only going to, I mean, people are

00:38:21.019 --> 00:38:22.239
going to invest in this if they're confident

00:38:22.239 --> 00:38:24.760
it can grow bottom line revenue. But you're going

00:38:24.760 --> 00:38:26.559
to want to be able to track this and measure

00:38:26.559 --> 00:38:29.280
this and show ROI. Like, yeah, we work really

00:38:29.280 --> 00:38:32.059
hard to make sure our products are on ChatGPT.

00:38:32.059 --> 00:38:34.019
And hey, look at how many purchases we have.

00:38:34.079 --> 00:38:35.559
That seems like a bit of a no brainer for them

00:38:35.559 --> 00:38:37.500
to build that in, doesn't it? It does. And I

00:38:37.500 --> 00:38:39.760
think that will come alongside the launch of

00:38:39.760 --> 00:38:43.599
ChatGPT ads, which this isn't a prediction, but

00:38:43.599 --> 00:38:46.940
they'll be coming very soon. You know what? That

00:38:46.940 --> 00:38:49.199
might just be my next prediction, Martin. But

00:38:49.199 --> 00:38:51.579
before we get to that prediction, let's do your

00:38:51.579 --> 00:38:55.559
next prediction. So my prediction had to change,

00:38:55.559 --> 00:38:58.179
actually. I had to revise my prediction because

00:38:58.179 --> 00:39:03.139
after I wrote the prediction, it came true within

00:39:03.139 --> 00:39:06.920
hours. I originally predicted that Anthropic

00:39:06.920 --> 00:39:11.619
would launch a mainstream desktop agent for knowledge

00:39:11.619 --> 00:39:14.989
workers. much like they have clawed code for

00:39:14.989 --> 00:39:19.050
programmers and software developers. Um, and

00:39:19.050 --> 00:39:23.469
I put that down in my prediction list and within

00:39:23.469 --> 00:39:27.329
a few hours of me writing that down, flawed cowork

00:39:27.329 --> 00:39:30.769
was announced, which is exactly that a, at the

00:39:30.769 --> 00:39:34.269
moment it's only available on Mac OS, but it

00:39:34.269 --> 00:39:40.289
is a flawed based agentic assistant that is capable

00:39:40.289 --> 00:39:44.019
of executing tasks. And it doesn't require you

00:39:44.019 --> 00:39:47.519
to use terminal or anything like an IDE if you're

00:39:47.519 --> 00:39:50.380
a programmer or a software developer. So if you

00:39:50.380 --> 00:39:53.159
haven't tried that yet, yeah, check it out. It's

00:39:53.159 --> 00:39:55.000
available for Mac users. And if you've got a

00:39:55.000 --> 00:39:58.139
Mac's plan on Claude, then go get stuck into

00:39:58.139 --> 00:40:00.840
that. But that is not my prediction. My revised

00:40:00.840 --> 00:40:05.099
prediction is associated to Anthropic and really

00:40:05.099 --> 00:40:06.820
comes off the back of that particular product

00:40:06.820 --> 00:40:10.340
launch. I predict Anthropic will launch a dedicated.

00:40:11.070 --> 00:40:17.329
agent plan for teams. So this is, I'm expecting

00:40:17.329 --> 00:40:20.730
them to release a separate subscription plan

00:40:20.730 --> 00:40:25.030
branded and marketed as an agent plan for teams

00:40:25.030 --> 00:40:27.929
or organizations distinct from their existing

00:40:27.929 --> 00:40:32.949
pro max or the team and enterprise plans already.

00:40:33.070 --> 00:40:37.090
And this is designed specifically around agentic

00:40:37.090 --> 00:40:42.349
task execution. And there will be metrics of

00:40:42.349 --> 00:40:46.389
things like how many tasks it can run, how many

00:40:46.389 --> 00:40:48.590
sub -agents you can run. And I think it will

00:40:48.590 --> 00:40:54.030
be really tied to a new product line. But that's

00:40:54.030 --> 00:40:58.130
what I expect. And busy marketing teams will

00:40:58.130 --> 00:41:01.690
be able to subscribe to this plan and there'll

00:41:01.690 --> 00:41:04.750
be ready -made templates out of the box. And

00:41:04.750 --> 00:41:07.289
we'll see this already coming through the likes

00:41:07.289 --> 00:41:11.079
of... flawed code and flawed co -work where people

00:41:11.079 --> 00:41:15.219
are creating skills documents um i think we'll

00:41:15.219 --> 00:41:18.039
see a similar thing but it'll be released under

00:41:18.039 --> 00:41:21.119
a new plan this is a super interesting one i'm

00:41:21.119 --> 00:41:25.559
kind of glad that um clawed co -work or whatever

00:41:25.559 --> 00:41:28.440
it was called got released because um i think

00:41:28.440 --> 00:41:30.739
this is a wonderful speculation especially because

00:41:32.460 --> 00:41:34.840
I think there's a lot more space here in terms

00:41:34.840 --> 00:41:37.280
of possibilities than meets the eye. And I give

00:41:37.280 --> 00:41:39.860
an example. Do you remember when we were watching

00:41:39.860 --> 00:41:44.900
a Google AI launch event? It was probably May

00:41:44.900 --> 00:41:47.860
this year. And they had an example of people

00:41:47.860 --> 00:41:51.699
collaborating with agents as co -workers in their

00:41:51.699 --> 00:41:56.320
Google chat platform. But I just sort of thought

00:41:56.320 --> 00:42:01.380
Slack, right? You'd Slack an AI co -worker. As

00:42:01.380 --> 00:42:03.840
if you were slacking a human co -worker, like,

00:42:03.900 --> 00:42:06.559
oh, where's this project? And then the co -worker

00:42:06.559 --> 00:42:09.519
would come back and let you know. So I wonder

00:42:09.519 --> 00:42:15.139
if we might see the ability to hire agentic co

00:42:15.139 --> 00:42:20.840
-workers. So whether there's an agent plan or

00:42:20.840 --> 00:42:25.840
you pay $500 a month and you get, I don't know,

00:42:25.900 --> 00:42:30.179
an intern that's capable of X, Y, Z. They can

00:42:30.179 --> 00:42:32.900
project manage like this. They can do research

00:42:32.900 --> 00:42:35.760
like this. And they can execute tasks like this.

00:42:35.820 --> 00:42:38.500
And you hire them like you would a person. And

00:42:38.500 --> 00:42:39.920
that's what you get. And that's the way they

00:42:39.920 --> 00:42:42.739
frame it. I hadn't really thought about that

00:42:42.739 --> 00:42:45.880
until you said what you said. I don't know. What

00:42:45.880 --> 00:42:47.340
do you think? How do you think that will play

00:42:47.340 --> 00:42:52.420
out? I think the market will be fumbling around

00:42:52.420 --> 00:42:55.400
to try to find the right pricing structure to

00:42:55.400 --> 00:42:58.420
begin with, most certainly. And I think every...

00:42:58.989 --> 00:43:03.210
one of those kinds of propositions and pricing

00:43:03.210 --> 00:43:07.489
structures will be tried um hire a team and it's

00:43:07.489 --> 00:43:10.230
the iron marketing team and they give you one

00:43:10.230 --> 00:43:13.809
coordinator or one strategist agent one coordinator

00:43:13.809 --> 00:43:19.190
and then a creative an editor aid media specialist

00:43:19.190 --> 00:43:22.289
whatever it may be and you pay two thousand dollars

00:43:22.289 --> 00:43:26.469
per month and you get this cluster um but i do

00:43:26.469 --> 00:43:28.760
think it's coming these organizations have got

00:43:28.760 --> 00:43:35.239
to to monetize they need to get more uh art well

00:43:35.239 --> 00:43:38.059
or should i say higher value subscriptions from

00:43:38.059 --> 00:43:41.119
companies they're the ones that are going to

00:43:41.119 --> 00:43:44.739
be paying it was a probably 18 months ago that

00:43:44.739 --> 00:43:48.920
we first heard chat gpt had spoken to investors

00:43:48.920 --> 00:43:52.719
and said they were preparing for a world where

00:43:52.719 --> 00:43:55.260
there is a $2 ,000 a month subscription and even

00:43:55.260 --> 00:43:58.920
a $20 ,000 a month subscription. And those haven't

00:43:58.920 --> 00:44:01.960
come to pass yet. We've got a $200 a month pro

00:44:01.960 --> 00:44:04.260
plan available across the board. You've got that

00:44:04.260 --> 00:44:07.519
in ChatGPT, Claude and Gem and I have all got

00:44:07.519 --> 00:44:11.019
equivalent plans around that value. Where's that

00:44:11.019 --> 00:44:13.539
$2 ,000 a month plan? And what's that doing?

00:44:14.039 --> 00:44:15.699
And I think this is the year that we'll see it,

00:44:15.760 --> 00:44:18.199
whatever shape or form that looks like, but I

00:44:18.199 --> 00:44:23.269
think it will be around you. by a team or an

00:44:23.269 --> 00:44:26.889
individual agent with sub -agents this is where

00:44:26.889 --> 00:44:28.869
we're going it's goldmine i think you're right

00:44:28.869 --> 00:44:31.110
i think they'll fumble around different companies

00:44:31.110 --> 00:44:33.969
and platforms will try different things to see

00:44:33.969 --> 00:44:39.010
what resonates with the market um yeah it's interesting

00:44:39.010 --> 00:44:41.690
what you mentioned about obviously we're both

00:44:41.690 --> 00:44:43.570
marketers so that's why you think about hiring

00:44:43.570 --> 00:44:46.929
a marketing team in that way i wonder if the

00:44:46.929 --> 00:44:50.750
reality would be Once you have an extremely capable

00:44:50.750 --> 00:44:55.150
agent, I wonder if it ends up like a chatbot.

00:44:55.269 --> 00:44:58.769
I think it was Bloomberg tried to fine -tune

00:44:58.769 --> 00:45:02.030
or even create their own model based on all of

00:45:02.030 --> 00:45:04.730
their available finance data. And then it turned

00:45:04.730 --> 00:45:07.130
out to not be better. I think it was based on

00:45:07.130 --> 00:45:09.190
GPT -3 and it turned out to not be better than

00:45:09.190 --> 00:45:11.610
GPT -4. So you did all that training and you

00:45:11.610 --> 00:45:13.929
end up with a product that's worse than the next

00:45:13.929 --> 00:45:16.329
iteration of models. So one question would be,

00:45:17.070 --> 00:45:20.550
If you get a really good agent, isn't it just

00:45:20.550 --> 00:45:25.570
the same agents doing all those jobs? And therefore,

00:45:25.650 --> 00:45:28.730
in theory, couldn't you just hire one agent?

00:45:29.349 --> 00:45:33.530
But, and I think to your point, in order to monetize

00:45:33.530 --> 00:45:37.590
this effectively, this is how these platforms

00:45:37.590 --> 00:45:39.989
need to think, right? They'll want you to hire

00:45:39.989 --> 00:45:45.130
five team members or, you know, or... or six

00:45:45.130 --> 00:45:47.550
different divisional capabilities or whatever

00:45:47.550 --> 00:45:49.809
so they can charge you six times instead of one

00:45:49.809 --> 00:45:52.989
time and i do think as we see the monetization

00:45:52.989 --> 00:45:56.349
of the models and the agents over time trying

00:45:56.349 --> 00:45:59.090
to build an ever bigger gap between the cost

00:45:59.090 --> 00:46:02.750
of the tokens to get stuff done and the outputs

00:46:02.750 --> 00:46:05.949
and, dare I say, outcomes offered by the agents

00:46:05.949 --> 00:46:08.469
in terms of what their commercial value is. Obviously,

00:46:08.489 --> 00:46:11.190
the platforms want that to be as big as possible.

00:46:11.369 --> 00:46:13.869
And I think we'll see that get stretched from

00:46:13.869 --> 00:46:17.530
both ends. I think token production for different

00:46:17.530 --> 00:46:21.190
types of tasks will drop. Maybe the systems will

00:46:21.190 --> 00:46:25.409
auto route to cheaper. faster models when the

00:46:25.409 --> 00:46:28.630
task allows for it but you as the consumer don't

00:46:28.630 --> 00:46:30.849
necessarily get the benefit of that but they

00:46:30.849 --> 00:46:33.349
just get to deliver cheaper at the same time

00:46:33.349 --> 00:46:36.030
as finding commercial outcomes that are worth

00:46:36.030 --> 00:46:39.050
a lot of value or money to a business and then

00:46:39.050 --> 00:46:41.130
charging based on the value not based on the

00:46:41.130 --> 00:46:43.909
tokens yeah absolutely they've got to position

00:46:43.909 --> 00:46:46.730
them not just as it's this price per million

00:46:46.730 --> 00:46:49.889
tokens in and per million tokens out they have

00:46:49.889 --> 00:46:53.539
to price it as hey look uh the value that we

00:46:53.539 --> 00:46:57.119
are delivering for your organization how we package

00:46:57.119 --> 00:46:59.260
that up and bundle it that's our secret sauce

00:46:59.260 --> 00:47:01.860
don't you worry about that but this to your point

00:47:01.860 --> 00:47:03.719
that was one of the things that i found really

00:47:03.719 --> 00:47:07.980
interesting when we saw the launch of flawed

00:47:07.980 --> 00:47:15.420
opus 4 .5 was its ability to achieve these frontier

00:47:15.420 --> 00:47:20.739
state -of -the -art benchmarks But its token

00:47:20.739 --> 00:47:23.360
efficiency compared to the previous iterations

00:47:23.360 --> 00:47:26.559
was vastly improved. Basically, it didn't have

00:47:26.559 --> 00:47:29.539
to think as much. Therefore, the actual cost

00:47:29.539 --> 00:47:34.800
of execution was far lower. And that's exactly

00:47:34.800 --> 00:47:37.079
what we're talking about, isn't it? The model

00:47:37.079 --> 00:47:40.440
providers want it to be as cheap to deliver as

00:47:40.440 --> 00:47:43.280
possible, but the value at the front end for

00:47:43.280 --> 00:47:47.400
you as a consumer. remains high and if they can

00:47:47.400 --> 00:47:50.539
squeeze that token use down while still saying

00:47:50.539 --> 00:47:52.400
look we've got this agent that will deliver amazingly

00:47:52.400 --> 00:47:55.619
it uses the frontier model that's where they're

00:47:55.619 --> 00:47:58.280
going to start putting in that that extra bit

00:47:58.280 --> 00:48:01.099
of margin yeah i mean i'm quite excited the rumors

00:48:01.099 --> 00:48:04.659
are floating around for deep seek version four

00:48:04.659 --> 00:48:06.519
and for listeners that you know i've been listening

00:48:06.519 --> 00:48:08.360
to the podcast or you know paying quite a lot

00:48:08.360 --> 00:48:11.639
of attention to ai you'll know that some very

00:48:11.639 --> 00:48:13.760
big companies like nvidia their share prices

00:48:13.760 --> 00:48:17.219
took a wobble when deep seek volume version 3

00:48:17.219 --> 00:48:21.219
was released out of china because it was an open

00:48:21.219 --> 00:48:23.780
source model that you could effectively use for

00:48:23.780 --> 00:48:26.820
free that had frontier model capabilities at

00:48:26.820 --> 00:48:29.320
the time it was right up there with the best

00:48:29.320 --> 00:48:34.059
chat gpt and and anthropic clawed models built

00:48:34.059 --> 00:48:39.210
basically with very um like low training costs

00:48:39.210 --> 00:48:42.349
slow information requirements and deep sea version

00:48:42.349 --> 00:48:44.349
4 is going to come out and there's a lot of rumors

00:48:44.349 --> 00:48:46.510
already swirling about how they're going to make

00:48:46.510 --> 00:48:49.389
that better and one of the i think when historians

00:48:49.389 --> 00:48:51.650
write about this period and i think they will

00:48:51.650 --> 00:48:54.690
and when we all watch the movie which we i suspect

00:48:54.690 --> 00:48:56.889
i thought the whole movie was going to be about

00:48:56.889 --> 00:48:59.929
when sam altman got fired for 10 minutes at open

00:48:59.929 --> 00:49:02.730
ai but i actually think this movie is now a 10

00:49:02.730 --> 00:49:05.429
part Netflix series because of all the different

00:49:05.429 --> 00:49:07.530
twists and turns that we've seen and I think

00:49:07.530 --> 00:49:11.090
we will see more of. But the fact that high -end

00:49:11.090 --> 00:49:14.750
NVIDIA chips were restricted to Chinese researchers

00:49:14.750 --> 00:49:17.989
and a bunch of these other factors that then

00:49:17.989 --> 00:49:21.369
pushed them to make innovations that they might

00:49:21.369 --> 00:49:22.849
not have had to have made if they'd have had

00:49:22.849 --> 00:49:26.110
access to the latest tech and the sort of trade

00:49:26.110 --> 00:49:28.150
wars that we've seen between China and the US.

00:49:29.400 --> 00:49:32.599
Then China taking a much more open source approach

00:49:32.599 --> 00:49:34.760
to AI. All of these things are all playing in

00:49:34.760 --> 00:49:37.320
together and they're all sort of compounding

00:49:37.320 --> 00:49:40.219
on each other. And I wouldn't be surprised if

00:49:40.219 --> 00:49:42.239
there's some really interesting stuff in DeepSeek

00:49:42.239 --> 00:49:45.119
version four that has some of the things you

00:49:45.119 --> 00:49:49.420
described by like a model that almost as a, I

00:49:49.420 --> 00:49:50.840
don't know what the word I'm looking for is like

00:49:50.840 --> 00:49:53.500
an emergent capability gets a better answer than

00:49:53.500 --> 00:49:55.710
any other model, but somehow doesn't. half the

00:49:55.710 --> 00:49:57.550
tokens than any of the other models would have

00:49:57.550 --> 00:50:00.570
done and distillation and all these other things

00:50:00.570 --> 00:50:02.250
i think yeah i think we're going to see more

00:50:02.250 --> 00:50:04.849
and more of that um and i think we're going to

00:50:04.849 --> 00:50:06.469
need it because we're all running if we've got

00:50:06.469 --> 00:50:09.389
agents running around after us doing all this

00:50:09.389 --> 00:50:13.250
stuff for us the turk all the token burn is like

00:50:13.250 --> 00:50:16.409
going to be absolutely insane if we think um

00:50:16.409 --> 00:50:20.429
the amount of ai intelligence required right

00:50:20.429 --> 00:50:23.929
now is a lot like it's nothing compared to everybody's

00:50:23.929 --> 00:50:26.739
got a hundred agents they don't know they've

00:50:26.739 --> 00:50:28.280
got a hundred agents they've just got a gemini

00:50:28.280 --> 00:50:29.699
assistant on their phone but they've basically

00:50:29.699 --> 00:50:31.820
got a hundred agents running after them all the

00:50:31.820 --> 00:50:34.139
time we're going to need all of those innovations

00:50:34.139 --> 00:50:39.780
anyway um nice one that's a cracker right my

00:50:39.780 --> 00:50:42.340
next prediction and this is going to be this

00:50:42.340 --> 00:50:46.900
finishes my uh ai chatbot holy trinity is that

00:50:46.900 --> 00:50:50.800
ads come to the ai chatbots It's been so many

00:50:50.800 --> 00:50:55.000
rumours through 2025 that this was coming. Supposedly,

00:50:55.000 --> 00:50:58.500
ChatGPT was very close to launching some sort

00:50:58.500 --> 00:51:00.679
of ad experience or certainly testing an ad experience

00:51:00.679 --> 00:51:04.119
towards the end of last year and opted not to

00:51:04.119 --> 00:51:07.780
let it out too much into the wild. We're going

00:51:07.780 --> 00:51:09.480
to see, I think we're going to see ads. We see

00:51:09.480 --> 00:51:11.079
all these people trying to make sure their brands

00:51:11.079 --> 00:51:14.739
show up in generative search answers, whether

00:51:14.739 --> 00:51:16.559
that's when you search on Google or whether you're

00:51:16.559 --> 00:51:21.150
searching on ChatGPT. And as we have seen, one

00:51:21.150 --> 00:51:23.150
of the great ways to shortcut making sure your

00:51:23.150 --> 00:51:26.170
brand shows up in those searches and in those

00:51:26.170 --> 00:51:29.409
conversations is to pay. And what a fantastic

00:51:29.409 --> 00:51:32.210
way for all of these platforms to make some of

00:51:32.210 --> 00:51:34.809
the money back, given the huge amount of costs

00:51:34.809 --> 00:51:38.610
associated with training and managing and inferencing

00:51:38.610 --> 00:51:42.130
models. You mentioned it earlier, Martin, seems

00:51:42.130 --> 00:51:44.309
like you're very much in the camp of ads are

00:51:44.309 --> 00:51:47.590
coming. Absolutely, 100%. In fact, it was reported

00:51:47.590 --> 00:51:51.250
last week, I think I saw that OpenAI is actually

00:51:51.250 --> 00:51:55.789
trialing their ad product internally with their

00:51:55.789 --> 00:51:58.289
staff. They're obviously trying to get a feel

00:51:58.289 --> 00:52:02.070
for the user experience and try to make it work

00:52:02.070 --> 00:52:04.750
in a way that is not particularly annoying for

00:52:04.750 --> 00:52:07.570
people. Because let's be honest, as we've discussed

00:52:07.570 --> 00:52:11.289
in the past on the show, ads change the user

00:52:11.289 --> 00:52:14.739
experience. in quite a fundamental way particularly

00:52:14.739 --> 00:52:17.340
when you're dealing with a chatbot that you go

00:52:17.340 --> 00:52:22.039
to for advice you look for feedback on the best

00:52:22.039 --> 00:52:25.579
product in this scenario and when you're giving

00:52:25.579 --> 00:52:30.380
your uh you're you're having communications with

00:52:30.380 --> 00:52:32.920
it in a way that is quite different to the way

00:52:32.920 --> 00:52:35.099
that you would with say a normal search engine

00:52:35.099 --> 00:52:37.880
and it knows an awful lot about you you have

00:52:37.880 --> 00:52:42.170
a different Trust threshold with that product.

00:52:42.349 --> 00:52:47.610
So they're gonna have to Move quite cautiously

00:52:47.610 --> 00:52:50.710
through this terrain. I feel the moment that

00:52:50.710 --> 00:52:58.030
the ads start to Make people trust The product

00:52:58.030 --> 00:53:02.190
less I think that will be a a huge kicker for

00:53:02.190 --> 00:53:05.079
it and I think They are aware of this. I've seen

00:53:05.079 --> 00:53:08.039
Sam Altman address this directly in podcasts

00:53:08.039 --> 00:53:12.300
and interviews in the past. But it's one thing

00:53:12.300 --> 00:53:18.179
being aware of it. It's another thing actually

00:53:18.179 --> 00:53:22.260
making an ad product, which A, doesn't destroy

00:53:22.260 --> 00:53:24.760
trust in the brand, but also B, does deliver

00:53:24.760 --> 00:53:28.980
the promise to advertisers. I do know it's spot

00:53:28.980 --> 00:53:31.719
on. I sort of flip back and forth on this. On

00:53:31.719 --> 00:53:34.920
the one hand, It feels like the stakes for getting

00:53:34.920 --> 00:53:38.940
this ad experience right when in an AI conversational

00:53:38.940 --> 00:53:42.440
format are much higher than they were for getting

00:53:42.440 --> 00:53:46.760
ads right in a search box, right? So for Google

00:53:46.760 --> 00:53:49.519
ads, Google search ads, for example, for the

00:53:49.519 --> 00:53:51.659
reason you said, right? Like people are developing

00:53:51.659 --> 00:53:54.940
quasi personal relationships with these chatbots.

00:53:55.400 --> 00:53:57.739
There's a level of trust that comes with the

00:53:57.739 --> 00:54:01.059
interactions that they just can't afford to ruin.

00:54:01.769 --> 00:54:03.849
through the ad experience. So they really do

00:54:03.849 --> 00:54:06.510
have to get that right. And I think that's why

00:54:06.510 --> 00:54:09.110
we're also seeing, you know, that people aren't,

00:54:09.110 --> 00:54:11.150
you know, platforms are experimenting with this

00:54:11.150 --> 00:54:12.630
already. You've mentioned, you know, some of

00:54:12.630 --> 00:54:14.889
the stuff that ChapGPT and OpenAI have been doing,

00:54:14.909 --> 00:54:18.030
but Perplexity has been having some experiments

00:54:18.030 --> 00:54:19.829
as well with things like sponsored follow -up

00:54:19.829 --> 00:54:23.489
questions and answers. Microsoft is looking at

00:54:23.489 --> 00:54:27.010
some co -pilot native ad formats like showroom

00:54:27.010 --> 00:54:30.110
ads that like live inside the conversation. They're

00:54:30.110 --> 00:54:33.010
not sort of on the side of the conversation and

00:54:33.010 --> 00:54:34.750
i think we're going to see the platforms have

00:54:34.750 --> 00:54:39.110
to experiment a lot to figure out what is an

00:54:39.110 --> 00:54:42.070
appropriate way to deliver a sponsored experience

00:54:42.070 --> 00:54:46.150
inside these platforms without just making people

00:54:46.150 --> 00:54:48.690
really mad because i think the stakes are higher

00:54:48.690 --> 00:54:51.710
and to that point actually i saw a really interesting

00:54:51.710 --> 00:54:54.469
business model it was i don't know the name of

00:54:54.469 --> 00:54:57.010
the company and it's going to bug me now but

00:54:57.010 --> 00:55:05.469
ad sponsored inference Where you Get So this

00:55:05.469 --> 00:55:09.690
is for people using it for development. They're

00:55:09.690 --> 00:55:14.030
actually selling access to the frontier models

00:55:14.030 --> 00:55:21.949
like Claude 4 .5 But they are Running out so

00:55:21.949 --> 00:55:25.570
they're giving you $300 a month worth of tokens

00:55:25.570 --> 00:55:29.309
at the moment in this sign up But then they are

00:55:29.309 --> 00:55:34.269
selling you ads over your token use and inference

00:55:34.269 --> 00:55:37.250
through their platform. Interesting. So if I've

00:55:37.250 --> 00:55:39.230
understood, I probably understood incorrectly,

00:55:39.489 --> 00:55:42.289
but the first thing that springs to mind is basically

00:55:42.289 --> 00:55:46.429
this is just another surface to show banner ads

00:55:46.429 --> 00:55:50.800
on and you can have your AI. chatbot cheaper

00:55:50.800 --> 00:55:53.480
if you're willing to have ads on the left and

00:55:53.480 --> 00:55:56.099
the right hand side of the search experience

00:55:56.099 --> 00:55:58.179
is that what you're suggesting and the ads are

00:55:58.179 --> 00:56:01.699
somehow contextually relevant but it's mostly

00:56:01.699 --> 00:56:03.820
so you don't have to pay a hundred dollars a

00:56:03.820 --> 00:56:07.500
month yeah that seems to be that seems to be

00:56:07.500 --> 00:56:09.280
the pitch at the moment and i think the people

00:56:09.280 --> 00:56:11.739
are going for them the audience is predominantly

00:56:11.739 --> 00:56:17.199
very techie uh developer types so no doubt the

00:56:17.199 --> 00:56:20.039
advertisers that are working with this particular

00:56:20.039 --> 00:56:24.840
brand offering this ad sponsored inference they

00:56:24.840 --> 00:56:27.539
are probably brands targeting that particular

00:56:27.539 --> 00:56:30.239
community i'm not sure the intricacies of it

00:56:30.239 --> 00:56:33.079
but it's an interesting model to to see start

00:56:33.079 --> 00:56:37.400
um launching to the market yeah well imagine

00:56:37.400 --> 00:56:41.360
something even like even i say worse it's hard

00:56:41.360 --> 00:56:43.360
for me to imagine that this ad experience is

00:56:43.360 --> 00:56:46.989
pleasant and anyway and i and i really think

00:56:46.989 --> 00:56:50.170
that people will clamor for an ad free experience

00:56:50.170 --> 00:56:53.030
that they're willing to pay for think five dollars

00:56:53.030 --> 00:56:55.409
a month for netflix with ads twenty dollars a

00:56:55.409 --> 00:56:59.949
month for netflix without ads um like spotify

00:56:59.949 --> 00:57:03.829
style you can't have on keep your conversation

00:57:03.829 --> 00:57:08.679
going unless you have this ad pop up in chat

00:57:08.679 --> 00:57:12.019
gpt for 15 seconds then you can click the x to

00:57:12.019 --> 00:57:14.539
get rid of it and then you can continue a conversation

00:57:14.539 --> 00:57:19.440
just like the mobile gaming watch this video

00:57:19.440 --> 00:57:21.599
oh my god i can't think of anything worse that

00:57:21.599 --> 00:57:25.019
would yeah well there's a reason i subscribe

00:57:25.019 --> 00:57:30.300
to to so many services and it's so i am so ad

00:57:30.300 --> 00:57:35.760
uh allergic i don't know i'm not sure That's

00:57:35.760 --> 00:57:38.420
a terrible way of phrasing it, but yeah, I will

00:57:38.420 --> 00:57:42.320
steer clear of ads in any way that I can. Imagine

00:57:42.320 --> 00:57:45.360
if the ad play, because I'm pretty sure this

00:57:45.360 --> 00:57:48.739
is Netflix's play too, if the ad play is just

00:57:48.739 --> 00:57:51.440
a way to make the ads so annoying that people

00:57:51.440 --> 00:57:54.639
are like, you know what, this is worth $30 a

00:57:54.639 --> 00:57:56.500
month to not have to part with this experience

00:57:56.500 --> 00:57:58.820
anymore. And they don't actually intend to make

00:57:58.820 --> 00:58:03.110
any money out of the ads. themselves they just

00:58:03.110 --> 00:58:05.389
want to push people to actually sign up for a

00:58:05.389 --> 00:58:07.650
paid version of the platform i mean that's another

00:58:07.650 --> 00:58:10.050
possibility i guess that's certainly the way

00:58:10.050 --> 00:58:12.510
that youtube seems to have gone and let me tell

00:58:12.510 --> 00:58:15.190
you that youtube premium subscription is the

00:58:15.190 --> 00:58:19.070
last subscription that i would ever give up you

00:58:19.070 --> 00:58:22.170
will claw that out of my cold dead hands you

00:58:22.170 --> 00:58:24.010
know i couldn't agree more you're trying to watch

00:58:24.010 --> 00:58:25.909
like a video and you're right into it and then

00:58:25.909 --> 00:58:29.119
without warning it just goes with volume that's

00:58:29.119 --> 00:58:31.300
like 30 higher than the actual thing you're watching

00:58:31.300 --> 00:58:35.539
usually an ai voice now as well doing the voiceover

00:58:35.539 --> 00:58:38.260
it turns out if i only click on that ad i can

00:58:38.260 --> 00:58:43.099
finally win at crypto um yeah so what does it

00:58:43.099 --> 00:58:46.739
mean for marketers on this one ultimately i think

00:58:46.739 --> 00:58:49.559
begin preparing for being able to buy your way

00:58:49.559 --> 00:58:53.340
into this generative answer experience keep a

00:58:53.340 --> 00:58:56.980
very close eye on it um but in the same way as

00:58:56.980 --> 00:59:00.440
there was a a juxtaposition between seo and ads

00:59:00.440 --> 00:59:03.119
i think that will be here as well i think if

00:59:03.119 --> 00:59:05.480
you can find that there are aspects of important

00:59:05.480 --> 00:59:08.480
conversations that you would want to be part

00:59:08.480 --> 00:59:10.400
of that for whatever reason your brand isn't

00:59:10.400 --> 00:59:12.039
getting mentioned you're going to have to pay

00:59:12.039 --> 00:59:14.440
to shortcut your way in there whilst constantly

00:59:14.440 --> 00:59:16.800
looking for optimization capabilities to try

00:59:16.800 --> 00:59:19.260
and make sure that your products and brand is

00:59:19.260 --> 00:59:22.880
the one that ai is mentioning let's call it natively

00:59:22.880 --> 00:59:27.179
or you know as the genuine organic result if

00:59:27.179 --> 00:59:31.059
you will And I think one of the things that could

00:59:31.059 --> 00:59:33.340
be interesting is how we bid for these placements.

00:59:33.380 --> 00:59:35.599
If it's banner ad style, I think it will work

00:59:35.599 --> 00:59:37.699
exactly like it does now to a certain extent.

00:59:38.119 --> 00:59:40.880
But if they really double down on trying to make

00:59:40.880 --> 00:59:42.900
sure the product recommendations are really good

00:59:42.900 --> 00:59:48.699
and natural and fit with what the person would

00:59:48.699 --> 00:59:52.039
genuinely want to buy, having great products

00:59:52.039 --> 00:59:55.440
is going to be important, right? And having...

00:59:56.840 --> 01:00:00.420
I would hope honest, authentic, easily machine

01:00:00.420 --> 01:00:04.000
readable, comprehensive product information that

01:00:04.000 --> 01:00:07.480
made the AI feel, in inverted commas, safe or

01:00:07.480 --> 01:00:10.300
accurate, making that recommendation to the human

01:00:10.300 --> 01:00:12.880
because they had enough information to base their

01:00:12.880 --> 01:00:16.059
decision on. So lots of content around case studies,

01:00:16.260 --> 01:00:21.039
data, and just basically supporting information

01:00:21.039 --> 01:00:23.360
that makes the AI feel like it's making a good

01:00:23.360 --> 01:00:25.500
recommendation is going to be critical. another

01:00:25.500 --> 01:00:27.699
thing i would say for advertisers thinking about

01:00:27.699 --> 01:00:33.000
this as a future play is um thinking back to

01:00:33.000 --> 01:00:35.639
the things that gary vee has often said about

01:00:35.639 --> 01:00:38.719
google ads early on and how undervalued they

01:00:38.719 --> 01:00:43.260
were that these will be uh places where there

01:00:43.260 --> 01:00:45.639
is some short -term arbitrage where you can get

01:00:45.639 --> 01:00:51.469
great value because there will be a slower shift

01:00:51.469 --> 01:00:53.929
from the competition as well. So I think there'll

01:00:53.929 --> 01:00:58.710
be a bit of quick wins for people that move onto

01:00:58.710 --> 01:01:02.090
these platforms and adopt early, although that

01:01:02.090 --> 01:01:05.789
gap will soon close. But yeah, don't be afraid

01:01:05.789 --> 01:01:09.210
to shift some of your media spend onto them when

01:01:09.210 --> 01:01:11.389
they initially come about. That's really interesting.

01:01:12.130 --> 01:01:15.690
I'm probably way too inside the system to...

01:01:15.929 --> 01:01:20.050
to know but my gut feeling is because of the

01:01:20.050 --> 01:01:23.769
gary vaynerchuks of the world i think a lot of

01:01:23.769 --> 01:01:26.829
marketers are nervous and are kind of annoyed

01:01:26.829 --> 01:01:29.530
and feel like probably they missed some of those

01:01:29.530 --> 01:01:31.730
early arbitrage opportunities whether it was

01:01:31.730 --> 01:01:35.050
google search ads sponsored content on social

01:01:35.050 --> 01:01:39.550
and everybody that i speak to seems way more

01:01:39.550 --> 01:01:42.630
primed for like knowing that this is going to

01:01:42.630 --> 01:01:45.480
be a thing thinking i need to get in there early

01:01:45.480 --> 01:01:49.119
while there's a lot of value to be had you speak

01:01:49.119 --> 01:01:52.440
to way more marketers than i do martin am i too

01:01:52.440 --> 01:01:54.519
inside the system is nobody really thinking about

01:01:54.519 --> 01:01:57.059
this or yeah i think you are to be completely

01:01:57.059 --> 01:02:02.980
honest um i think there will be a six month one

01:02:02.980 --> 01:02:05.900
year two year window and then part of that will

01:02:05.900 --> 01:02:09.500
come about with if people are happy with their

01:02:09.500 --> 01:02:11.380
media spend at the moment they'll be reluctant

01:02:11.380 --> 01:02:15.130
to to reallocate budget i think there's just

01:02:15.130 --> 01:02:18.010
going to be a slow shift agencies that are delivering

01:02:18.010 --> 01:02:20.269
this are going to have to be figuring out how

01:02:20.269 --> 01:02:24.969
to use the systems so yeah i just think there'll

01:02:24.969 --> 01:02:29.409
be there'll be a period where might be a good

01:02:29.409 --> 01:02:33.130
place to to play in the paid media space that

01:02:33.130 --> 01:02:35.630
is a very fair point i'm also not particularly

01:02:35.630 --> 01:02:39.070
risk averse when it comes to speculating on new

01:02:39.070 --> 01:02:43.110
channels um and It's very easy to forget all

01:02:43.110 --> 01:02:45.650
of the new channels that people speculated on

01:02:45.650 --> 01:02:47.489
and invested time in that came to nothing. What

01:02:47.489 --> 01:02:49.769
was that one? There was one at the start of the

01:02:49.769 --> 01:02:51.570
pandemic. I can't remember what it was called,

01:02:51.650 --> 01:02:53.690
where basically it was like you could dial into

01:02:53.690 --> 01:02:55.369
a call that everybody was on. Do you remember

01:02:55.369 --> 01:03:00.130
what that was called? It was something house,

01:03:00.349 --> 01:03:02.070
wasn't it? Yeah, something like that. And everyone

01:03:02.070 --> 01:03:03.789
was like, yeah, this is the new channel which

01:03:03.789 --> 01:03:06.269
everyone's got to be on. And it was like super

01:03:06.269 --> 01:03:10.369
hot for like two months. And then it was... disappeared

01:03:10.369 --> 01:03:13.550
into the ether so yeah you can double down on

01:03:13.550 --> 01:03:17.570
channels and not win it does happen reasonably

01:03:17.570 --> 01:03:19.809
frequently as well so yeah i guess maybe that

01:03:19.809 --> 01:03:22.250
could put people off as well um i don't think

01:03:22.250 --> 01:03:23.809
this is going to be one of those to be honest

01:03:23.809 --> 01:03:26.909
but people might feel like that um let's go on

01:03:26.909 --> 01:03:28.969
to your next prediction then mine following on

01:03:28.969 --> 01:03:31.469
from what you said earlier about this is not

01:03:31.469 --> 01:03:36.630
a particularly newsy episode but also there was

01:03:36.630 --> 01:03:39.730
some news in December that was very relevant

01:03:39.730 --> 01:03:44.650
to this next prediction, Meta acquired my favorite

01:03:44.650 --> 01:03:48.530
agentic tool, Manus. Manus is a tool that I've

01:03:48.530 --> 01:03:52.030
been using for, well, for months now. I've got

01:03:52.030 --> 01:03:54.190
the subscription. I've been stress testing it.

01:03:54.210 --> 01:03:56.969
I've made some great little fun apps. I've had

01:03:56.969 --> 01:03:59.550
it execute entire workflows for me. I've been

01:03:59.550 --> 01:04:02.349
incredibly frustrated with it at times. I just

01:04:02.349 --> 01:04:04.210
started to feel like I was getting the hang of

01:04:04.210 --> 01:04:09.969
how to instruct it specify tasks to it and all

01:04:09.969 --> 01:04:12.630
of that kind of stuff and then meta came along

01:04:12.630 --> 01:04:15.190
and gobbled it up although importantly they do

01:04:15.190 --> 01:04:17.409
say they are going to run it as a standalone

01:04:17.409 --> 01:04:22.909
app but this acquisition did make me think about

01:04:22.909 --> 01:04:28.349
the future of the meta marketing suite so with

01:04:28.349 --> 01:04:32.050
that in mind my prediction is meta will launch

01:04:32.050 --> 01:04:40.269
an agentic Support agent via messenger or WhatsApp.

01:04:40.590 --> 01:04:47.469
And this will be a customer facing business interfacing

01:04:47.469 --> 01:04:51.329
agent that will be able to plug and play with

01:04:51.329 --> 01:04:56.789
almost any platform that your business operates

01:04:56.789 --> 01:05:01.090
with. So I'm expecting the computer use functionality

01:05:01.090 --> 01:05:06.159
to be used from Manus. they'll basically use

01:05:06.159 --> 01:05:10.980
this technology stack to basically offer to customers

01:05:10.980 --> 01:05:14.659
and to small business owners a customer support

01:05:14.659 --> 01:05:17.719
agent or whatever it may be and it will be able

01:05:17.719 --> 01:05:20.579
to go on to the websites and answer questions

01:05:20.579 --> 01:05:22.940
you'll be able to upload knowledge documents

01:05:22.940 --> 01:05:27.639
and it'll be powered in the back end by the new

01:05:27.639 --> 01:05:31.619
iteration of the llama series of models or whatever

01:05:31.619 --> 01:05:34.670
the the next iteration is in the closed source

01:05:34.670 --> 01:05:37.750
future of large language models from meta but

01:05:37.750 --> 01:05:42.809
yeah i expect them to to have a bot that is an

01:05:42.809 --> 01:05:46.090
official meta product it can take actions so

01:05:46.090 --> 01:05:50.369
order lookups returned bookings not just through

01:05:50.369 --> 01:05:54.210
official partner integrations but i expect it

01:05:54.210 --> 01:05:57.010
to go a bit beyond that and it can hand off to

01:05:57.010 --> 01:06:01.320
humans while maintaining context Yeah, that's

01:06:01.320 --> 01:06:03.860
an interesting one. I mean, I assume they have

01:06:03.860 --> 01:06:06.780
a strategy, but I think probably Mark's strategy

01:06:06.780 --> 01:06:10.420
is we'd love to have Martin as a customer. So

01:06:10.420 --> 01:06:13.639
let's buy the company that issued his pendant.

01:06:14.400 --> 01:06:18.079
Let's buy his favorite agent because he spends

01:06:18.079 --> 01:06:20.199
so much money on lifetime deals. Like we need

01:06:20.199 --> 01:06:23.800
this guy to be. you know, an active customer

01:06:23.800 --> 01:06:26.139
in, and he's got a very famous podcast as well

01:06:26.139 --> 01:06:28.800
about AI and marketing. So I think they're just

01:06:28.800 --> 01:06:30.739
coming after you, but it's a bit of a shame for

01:06:30.739 --> 01:06:33.019
you to have another one of your things bought

01:06:33.019 --> 01:06:37.320
up by the least trustworthy company in many people's

01:06:37.320 --> 01:06:40.159
eyes, allegedly, et cetera, et cetera. Yeah.

01:06:40.260 --> 01:06:42.360
So sorry about that, Martin, that you've had

01:06:42.360 --> 01:06:43.820
to go through that again, but hopefully you'll

01:06:43.820 --> 01:06:45.760
still have good access to Manus to try and do

01:06:45.760 --> 01:06:48.000
the things that you want to do. And super interesting

01:06:48.000 --> 01:06:50.920
that. Because it will bring those capabilities

01:06:50.920 --> 01:06:53.400
more quickly to people who wouldn't be willing

01:06:53.400 --> 01:06:56.000
to go play with a Manus. I think they'll be able

01:06:56.000 --> 01:06:59.960
to turn it into a easy to use interface. They'll

01:06:59.960 --> 01:07:02.820
be able to productize it. They've just got so

01:07:02.820 --> 01:07:06.739
many more heads available to think about that

01:07:06.739 --> 01:07:11.039
rather than a small team of engineers thinking

01:07:11.039 --> 01:07:12.900
about how to make the thing work. They'll be

01:07:12.900 --> 01:07:16.199
able to plug it into their existing. ai suite

01:07:16.199 --> 01:07:20.239
you know they've got so many ways ai is powering

01:07:20.239 --> 01:07:23.920
the advertising platform but when they can then

01:07:23.920 --> 01:07:27.360
plug this into the interfaces where people are

01:07:27.360 --> 01:07:31.800
already chatting so whatsapp and messenger i

01:07:31.800 --> 01:07:33.880
think it's just going to be a huge amount of

01:07:33.880 --> 01:07:38.900
unlock um for end customers they'll be able to

01:07:38.900 --> 01:07:41.340
productize this agentic experience and in some

01:07:41.340 --> 01:07:46.000
ways what i'm describing here is What I expect

01:07:46.000 --> 01:07:51.840
from Google with a productized agentic suite

01:07:51.840 --> 01:07:57.159
and flawed with an agent plan that we talked

01:07:57.159 --> 01:08:00.780
about earlier, I just think this will be bundled

01:08:00.780 --> 01:08:03.840
into their existing product offering. And the

01:08:03.840 --> 01:08:06.119
way that they'll monetize it will be through

01:08:06.119 --> 01:08:08.760
advertising because you'll be spending more money.

01:08:08.840 --> 01:08:11.000
They won't be charging on token costs. They know

01:08:11.000 --> 01:08:13.760
that it will be based on spend more advertising.

01:08:14.590 --> 01:08:18.569
across our suite of platforms where people are

01:08:18.569 --> 01:08:23.710
already hanging out and uh yeah we'll manage

01:08:23.710 --> 01:08:26.130
the transactions we'll manage the customer service

01:08:26.130 --> 01:08:29.010
we'll manage the sales process don't worry about

01:08:29.010 --> 01:08:31.590
that hand off to us and we'll do all the integrations

01:08:31.590 --> 01:08:33.619
in the back end as well You could see the attraction

01:08:33.619 --> 01:08:35.680
of a platform like that. I think you're right.

01:08:35.760 --> 01:08:38.600
I know we're going to leave our big agent discussion

01:08:38.600 --> 01:08:41.760
to the end because 2026 being the year of agents,

01:08:41.819 --> 01:08:43.779
as it almost certainly will be. But seeing as

01:08:43.779 --> 01:08:47.039
we're on this topic of Manus, I think you're

01:08:47.039 --> 01:08:49.239
absolutely right that this has to be productized

01:08:49.239 --> 01:08:53.420
and brought to market in a way that sort of early

01:08:53.420 --> 01:08:56.640
adopter hobbyists, early adopter hobbyists are

01:08:56.640 --> 01:08:58.560
willing to take a certain approach, but most

01:08:58.560 --> 01:09:00.779
people are not going to be. And to bring this

01:09:00.779 --> 01:09:04.579
to life. I tried a new platform this week called

01:09:04.579 --> 01:09:07.000
Do Anything. It's basically like Manus. It's

01:09:07.000 --> 01:09:10.220
a chatbot. You give it a big picture outcome

01:09:10.220 --> 01:09:12.560
that you're trying to achieve rather than like

01:09:12.560 --> 01:09:15.880
ask it a question. And then it comes up with

01:09:15.880 --> 01:09:18.539
a plan. It spins up some sub -agents that can

01:09:18.539 --> 01:09:20.680
do different things. And then it goes off and

01:09:20.680 --> 01:09:23.819
it tries to execute your plan. $20 a month. I

01:09:23.819 --> 01:09:26.239
thought I'd give it a try for a month. My $20

01:09:26.239 --> 01:09:31.920
of credit lasted for 55 minutes. The outcome

01:09:31.920 --> 01:09:35.020
that I chose was one of the stock outcomes that

01:09:35.020 --> 01:09:38.899
it suggests that you try, which was to create

01:09:38.899 --> 01:09:43.460
a monetizable newsletter that has 2000 subscribers.

01:09:44.819 --> 01:09:46.819
So the first part of the process was kind of

01:09:46.819 --> 01:09:48.939
fun. It was like, here are some topics that are

01:09:48.939 --> 01:09:51.399
trending right now. Which of these are most interesting

01:09:51.399 --> 01:09:52.939
that you might want to start a newsletter on

01:09:52.939 --> 01:09:56.539
or suggest your own? So I chose something. Then

01:09:56.539 --> 01:09:58.380
it came up with its plan for how it was going

01:09:58.380 --> 01:10:03.619
to execute. It tried to create a, I think the

01:10:03.619 --> 01:10:05.939
way to pronounce it is Beehive, but it's one

01:10:05.939 --> 01:10:08.880
of the newsletter platforms like Substack. It

01:10:08.880 --> 01:10:11.840
went to try to create an account for me and just

01:10:11.840 --> 01:10:14.979
got stuck and just was burning credits trying

01:10:14.979 --> 01:10:18.899
to help me help it set up an account. And all

01:10:18.899 --> 01:10:21.039
I could see was it trying the same thing again

01:10:21.039 --> 01:10:22.859
and again and again. And I'd be like, should

01:10:22.859 --> 01:10:25.359
we take a step back and have a think about whether

01:10:25.359 --> 01:10:27.359
this is working or not? Which it did try and

01:10:27.359 --> 01:10:30.920
do several times. But one of its... ways of getting

01:10:30.920 --> 01:10:34.960
around this frustration was to go execute some

01:10:34.960 --> 01:10:36.880
of the other steps in the plan that i would have

01:10:36.880 --> 01:10:39.720
preferred come later so it started creating all

01:10:39.720 --> 01:10:42.920
these lead magnets it came up with this outreach

01:10:42.920 --> 01:10:46.000
strategy to bloggers that it was going to try

01:10:46.000 --> 01:10:48.060
and get them to subscribe to our newsletter by

01:10:48.060 --> 01:10:50.979
promising to subscribe to theirs or some other

01:10:50.979 --> 01:10:54.600
strategy which if I can't even get the newsletter

01:10:54.600 --> 01:10:56.619
live because the system doesn't work, I certainly

01:10:56.619 --> 01:10:59.039
don't need all that stuff. By the way, this is

01:10:59.039 --> 01:11:00.899
consuming credits like mad in the background.

01:11:01.640 --> 01:11:07.340
And it started sending pitch emails. Yeah, I

01:11:07.340 --> 01:11:10.220
didn't get to check the pitch emails because

01:11:10.220 --> 01:11:12.899
part of its value prop that makes it slightly

01:11:12.899 --> 01:11:16.119
different from Manus is it has its own email

01:11:16.119 --> 01:11:19.520
address and you get to choose its name like it's

01:11:19.520 --> 01:11:22.579
a person. So this newsletter is not by me. It's

01:11:22.579 --> 01:11:26.399
by, I had this crazy name, like Carlton Nightingale

01:11:26.399 --> 01:11:28.119
or something like that. Because it gave me options

01:11:28.119 --> 01:11:31.319
and you just keep hitting next and it just gets

01:11:31.319 --> 01:11:33.199
more and more creative the more times you hit

01:11:33.199 --> 01:11:36.840
next. And there was something about Carlton from

01:11:36.840 --> 01:11:40.819
Fresh Prince with the idea of having like Nightingale.

01:11:40.880 --> 01:11:42.920
That just seemed right to me. Didn't make it

01:11:42.920 --> 01:11:44.720
more capable though. So yeah, I started pitching.

01:11:45.340 --> 01:11:47.239
And I was like, you're pitching a newsletter

01:11:47.239 --> 01:11:49.739
that doesn't exist because you can't even log

01:11:49.739 --> 01:11:51.539
into the system and create the newsletter, even

01:11:51.539 --> 01:11:55.039
though this is the stock thing that they recommend

01:11:55.039 --> 01:11:57.859
that you try. So, yeah, I didn't invest in any

01:11:57.859 --> 01:11:59.420
more credits. And the reason for this story,

01:11:59.500 --> 01:12:03.539
Martin, is that is a hobbyist's playground. That

01:12:03.539 --> 01:12:06.380
is a not a commercially ready approach by any

01:12:06.380 --> 01:12:09.680
stretch of the imagination. And so to your point,

01:12:09.699 --> 01:12:11.560
they have to productize that in a way that people

01:12:11.560 --> 01:12:14.020
can actually use it. And then the second point

01:12:14.020 --> 01:12:18.279
is when this gets good enough, my goodness, are

01:12:18.279 --> 01:12:22.039
we going to get deluged with crap content? My

01:12:22.039 --> 01:12:25.579
goodness. Because if you imagine, so the whole

01:12:25.579 --> 01:12:29.399
value proposition here is I should have come

01:12:29.399 --> 01:12:35.779
back and I had 2000 subscribers. So 500 odd people

01:12:35.779 --> 01:12:38.100
who really didn't deserve to get pitched by this

01:12:38.100 --> 01:12:43.380
bot would have got pitched. and a load of content

01:12:43.380 --> 01:12:46.000
about whatever the topic was that added probably

01:12:46.000 --> 01:12:48.020
no unique value because it was a regurgitation

01:12:48.020 --> 01:12:50.079
of stuff that's already available on the web.

01:12:50.119 --> 01:12:51.920
It's getting sent out to people. By the way,

01:12:51.939 --> 01:12:54.840
not accurate, probably. So because I'm interested

01:12:54.840 --> 01:12:57.340
in science, it was a science -based one. What

01:12:57.340 --> 01:12:59.859
if it's telling people absolute tosh? That's

01:12:59.859 --> 01:13:06.359
incorrect. Yeah, so I think there's quite a sizable

01:13:06.359 --> 01:13:10.100
gap to traverse before this prediction's possible.

01:13:10.670 --> 01:13:14.130
Well, I think that goes to, well, first of all,

01:13:14.130 --> 01:13:16.229
Paul, actually, let me just say, I wish I hadn't

01:13:16.229 --> 01:13:17.310
told that story because you're going to make

01:13:17.310 --> 01:13:20.449
people think that agents aren't very good. And

01:13:20.449 --> 01:13:22.029
actually some of them are really good. And we

01:13:22.029 --> 01:13:24.750
need people to understand that to make agents

01:13:24.750 --> 01:13:26.229
really good, you need to know how to specify

01:13:26.229 --> 01:13:28.510
correctly. And they can do that by speaking to

01:13:28.510 --> 01:13:33.270
the likes of me and you. Correct. Not me. But

01:13:33.270 --> 01:13:34.750
this was just an experiment. Come on, people,

01:13:34.770 --> 01:13:36.510
it was just an experiment. The system promised.

01:13:36.729 --> 01:13:38.310
And that's why I was like, well, show us what

01:13:38.310 --> 01:13:41.899
you can do then. that does go to the the point

01:13:41.899 --> 01:13:44.340
about you know it's ai slop it's bots speaking

01:13:44.340 --> 01:13:47.439
to bots it's bots pitching to bots that are signing

01:13:47.439 --> 01:13:50.680
up to bots reading bots regurgitating bots it's

01:13:50.680 --> 01:13:54.779
just a bot hellscape but taking it back to to

01:13:54.779 --> 01:13:58.659
meta just for a second and i can see a world

01:13:58.659 --> 01:14:04.399
where that even with the customer service side

01:14:04.399 --> 01:14:09.859
of things like having seen how how Manus actually

01:14:09.859 --> 01:14:12.159
works and having experienced some of the major

01:14:12.159 --> 01:14:15.239
frustrations of trying to get things all working

01:14:15.239 --> 01:14:17.659
inside the Meta platform, particularly when you're

01:14:17.659 --> 01:14:21.479
dealing with myriad different e -commerce platforms,

01:14:21.680 --> 01:14:24.340
CRM platforms, marketing automation platforms,

01:14:24.659 --> 01:14:28.239
lots of small business owners trying to plug

01:14:28.239 --> 01:14:31.159
together all of these different bits is really

01:14:31.159 --> 01:14:35.020
hard. Whereas I can see a world where Meta actually

01:14:35.020 --> 01:14:40.340
deploys the Manus bot as a do it for you agent

01:14:40.340 --> 01:14:44.899
you know we will you tell us your uh your login

01:14:44.899 --> 01:14:48.340
for your crm or you take over the computer use

01:14:48.340 --> 01:14:51.640
and we'll hook up your meta advertising platform

01:14:51.640 --> 01:14:54.800
to your shopify account and make sure that all

01:14:54.800 --> 01:14:57.539
the reporting is correct and we'll do that for

01:14:57.539 --> 01:15:02.020
you and they're not using a team out in bangladesh

01:15:02.020 --> 01:15:04.600
or whatever it is that's running their customer

01:15:04.600 --> 01:15:07.930
service they're using empowering it with the

01:15:07.930 --> 01:15:12.270
Manus bot agent. Yeah, I think the interesting

01:15:12.270 --> 01:15:14.770
thing about the value prop of getting this do

01:15:14.770 --> 01:15:18.250
anything app to do that was it kind of misses

01:15:18.250 --> 01:15:20.029
the point of how this will really need to be

01:15:20.029 --> 01:15:25.810
rolled out, which is human in the loop. It's

01:15:25.810 --> 01:15:31.010
automation and autonomy in low risk areas that

01:15:31.010 --> 01:15:33.649
don't have a big impact. That's why I was like,

01:15:33.789 --> 01:15:36.920
why have you just sent a pitch email? what are

01:15:36.920 --> 01:15:41.180
you doing like if i hired you if i hired a junior

01:15:41.180 --> 01:15:43.359
and they started sending pitch emails about running

01:15:43.359 --> 01:15:45.520
them through me they'd be hell to pay like i'm

01:15:45.520 --> 01:15:48.840
not going to let an ai hallucinating bot do it

01:15:48.840 --> 01:15:54.420
um yeah so that's not how you deploy these things

01:15:54.420 --> 01:15:58.319
you find those use cases you you carve them up

01:15:58.319 --> 01:16:01.359
and you and you stress test to make and you start

01:16:01.359 --> 01:16:04.920
in the safe places where a mistake is not going

01:16:04.920 --> 01:16:08.109
to be have that you know too big of ramifications

01:16:08.109 --> 01:16:10.010
and i think you're right that's exactly how they

01:16:10.010 --> 01:16:12.670
will then they will then bring that offering

01:16:12.670 --> 01:16:15.970
to people and it will have to evolve and it will

01:16:15.970 --> 01:16:20.310
have to mature and i would be extremely skeptical

01:16:20.310 --> 01:16:24.409
of any platform that went from where we are now

01:16:24.409 --> 01:16:28.270
to describe the business you want and check your

01:16:28.270 --> 01:16:32.130
bank account in a month right like no not going

01:16:32.130 --> 01:16:34.869
to be like that Right, in the interest of time,

01:16:34.930 --> 01:16:37.270
because we've had our listeners on for a while

01:16:37.270 --> 01:16:42.109
now, I'll move on to my next prediction. But

01:16:42.109 --> 01:16:43.710
I'm going to speed this up and go through the

01:16:43.710 --> 01:16:47.430
quick version, Martin, and then maybe let's talk

01:16:47.430 --> 01:16:50.609
about agents a little bit at the end. So a couple

01:16:50.609 --> 01:16:52.689
of other quick predictions that I had was that

01:16:52.689 --> 01:16:55.970
I think AI memory is going to get solved and

01:16:55.970 --> 01:16:59.829
that persistent personalisation becomes the default,

01:16:59.890 --> 01:17:02.020
one of the big problems with... AI models right

01:17:02.020 --> 01:17:03.739
now is you start a chat and you're starting from

01:17:03.739 --> 01:17:05.180
fresh. It doesn't know anything about you. It

01:17:05.180 --> 01:17:06.460
doesn't remember anything else that you said.

01:17:07.020 --> 01:17:09.439
Everything's based on stuffing a load of stuff

01:17:09.439 --> 01:17:12.579
into the context window, which has pros and cons.

01:17:12.720 --> 01:17:16.100
It's extremely compute intense to do that, burns

01:17:16.100 --> 01:17:19.800
more tokens. And the agents are only, sorry,

01:17:20.119 --> 01:17:22.880
the chatbots are only so, so necessarily pulling

01:17:22.880 --> 01:17:25.119
the right thing from their context at the right

01:17:25.119 --> 01:17:28.739
time. As an aside, I'm seeing a lot of the moment

01:17:28.739 --> 01:17:34.020
of when I create, like a custom gpt or a gem

01:17:34.020 --> 01:17:37.560
or even a project the obsession of the chatbot

01:17:37.560 --> 01:17:40.199
to use every single thing in its context in the

01:17:40.199 --> 01:17:42.880
answer it's like the user gave this to me it

01:17:42.880 --> 01:17:45.680
must be important it's like you've got 20 documents

01:17:45.680 --> 01:17:47.739
this question is actually only about one of the

01:17:47.739 --> 01:17:50.699
documents be smart enough to realize you don't

01:17:50.699 --> 01:17:52.659
have to reference what's in all the other documents

01:17:52.659 --> 01:17:54.520
it's not relevant anyway this is supposed to

01:17:54.520 --> 01:17:57.840
be the quick one um but it's been some interesting

01:17:57.840 --> 01:18:02.020
newsy stuff this week because There was a couple

01:18:02.020 --> 01:18:05.180
of papers, one of which looks like a new capability

01:18:05.180 --> 01:18:07.739
for memory that we might see in DeepSeek version

01:18:07.739 --> 01:18:10.619
4, which is interesting. And then MIT released

01:18:10.619 --> 01:18:13.439
some research for a new memory capability. We

01:18:13.439 --> 01:18:15.020
know the labs are all over this at the moment,

01:18:15.079 --> 01:18:17.039
but obviously it becomes a fairly massive game

01:18:17.039 --> 01:18:20.760
changer for lots of use cases when AI can start

01:18:20.760 --> 01:18:24.960
to remember things and access memories in a much

01:18:24.960 --> 01:18:27.020
better way than it does now, both in terms of

01:18:27.020 --> 01:18:29.770
as a consumer. capturing all of this insight

01:18:29.770 --> 01:18:31.510
about you over time that Martin and I have talked

01:18:31.510 --> 01:18:33.329
throughout the podcast, but also as a business

01:18:33.329 --> 01:18:35.949
user, not have to constantly explain certain

01:18:35.949 --> 01:18:37.829
things about your products, your customers, your

01:18:37.829 --> 01:18:40.250
business, and just having that information available

01:18:40.250 --> 01:18:44.689
and accessible in a way that the chatbot remembers

01:18:44.689 --> 01:18:48.510
in a smart way. So I think we'll see memory get

01:18:48.510 --> 01:18:51.149
solved. Thoughts on memory, Martin? It's the

01:18:51.149 --> 01:18:53.710
missing piece of the puzzle. And some people

01:18:53.710 --> 01:18:57.289
argue that cracking that is basically the key

01:18:57.289 --> 01:19:01.279
to... cracking AGI. So yeah, if they get there,

01:19:01.420 --> 01:19:03.939
that will be quite something. In fact, it's interesting

01:19:03.939 --> 01:19:07.619
you mentioned about memory and the use of it.

01:19:07.699 --> 01:19:13.380
I find as much as ChatGPT, I don't think it is

01:19:13.380 --> 01:19:19.779
the best model. I prefer Claude, particularly

01:19:19.779 --> 01:19:23.340
for professional tasks. So if I'm doing stuff

01:19:23.340 --> 01:19:26.600
in CRM or if I'm coding a little thing together.

01:19:27.149 --> 01:19:30.710
very much prefer Claude, but my daily driver,

01:19:30.850 --> 01:19:33.970
the thing I turn to, the thing that is on my

01:19:33.970 --> 01:19:37.050
second screen at all times is ChatGPT. And why

01:19:37.050 --> 01:19:40.409
is that? I think it handles memory and personalization

01:19:40.409 --> 01:19:43.869
better than any of the others at this moment

01:19:43.869 --> 01:19:47.050
in time. Case in point, yesterday I put a prompt

01:19:47.050 --> 01:19:52.489
into it and it referenced something that I haven't

01:19:52.489 --> 01:19:54.630
given it in its custom instructions. It referenced

01:19:54.630 --> 01:19:57.260
something from a... from a chat that I'd had

01:19:57.260 --> 01:20:00.819
with it a long time ago about clients I'd worked

01:20:00.819 --> 01:20:05.020
with over the previous six months. And it pulled

01:20:05.020 --> 01:20:06.939
that information into the chat because it was

01:20:06.939 --> 01:20:11.520
able to search across. Now that's using the RAG,

01:20:11.520 --> 01:20:15.199
the Retrieval Augmented Generation method of

01:20:15.199 --> 01:20:19.119
managing memory. But it did a really nice job

01:20:19.119 --> 01:20:22.039
of it. And I can see that when they do correct

01:20:22.039 --> 01:20:24.699
this infinite memory piece, not only will it

01:20:24.699 --> 01:20:27.159
be good for those kinds of contextual understanding,

01:20:27.460 --> 01:20:34.279
but it will facilitate learning that's beyond

01:20:34.279 --> 01:20:38.939
just kind of in context learning. Great. I mean,

01:20:38.960 --> 01:20:42.140
that unlocks so many self -training capabilities

01:20:42.140 --> 01:20:45.640
of AI as well. I mean, we could go deep down

01:20:45.640 --> 01:20:48.420
the AI rabbit hole here, but I think you're right

01:20:48.420 --> 01:20:51.340
as a user. And I think, what does this mean to

01:20:51.340 --> 01:20:55.880
marketers? This is where personalization at scale

01:20:55.880 --> 01:20:59.000
in your marketing becomes possible, is if platforms

01:20:59.000 --> 01:21:02.659
understand your target customers in that level

01:21:02.659 --> 01:21:05.359
of detail. Some critical questions that come

01:21:05.359 --> 01:21:08.399
up is, how can you capture as much information

01:21:08.399 --> 01:21:10.260
as your customers as possible? We talked about

01:21:10.260 --> 01:21:12.140
PayPal trying to get some of that, but how can

01:21:12.140 --> 01:21:17.279
you do that? Will the chatbots sell or rent this

01:21:17.279 --> 01:21:20.100
information in some form to brands as part of

01:21:20.100 --> 01:21:23.729
their paid offering? You don't know who all of

01:21:23.729 --> 01:21:26.109
the 20 people who would like your product are,

01:21:26.189 --> 01:21:28.869
but we do. And that's actually what becomes part

01:21:28.869 --> 01:21:31.350
of serving relevant ads to people is just having

01:21:31.350 --> 01:21:36.289
that crazily deep understanding of them. So,

01:21:36.409 --> 01:21:38.970
you know, we've talked a little bit in the past

01:21:38.970 --> 01:21:40.630
about how important it is for brands to really

01:21:40.630 --> 01:21:43.630
focus on their first party data that they're

01:21:43.630 --> 01:21:45.210
capturing on customers and how they might use

01:21:45.210 --> 01:21:47.189
that. Because the other side of this will be

01:21:47.189 --> 01:21:49.710
then using AI to execute hyper -personalised

01:21:49.710 --> 01:21:51.810
campaigns at scale in a way that humans couldn't

01:21:51.810 --> 01:21:54.189
do. So there's those sort of two parts of the

01:21:54.189 --> 01:21:56.970
pie there, really. But yeah, I just feel like

01:21:56.970 --> 01:21:58.550
memory, as you said, is going to be one of those

01:21:58.550 --> 01:22:00.489
critical pieces. And I feel like it's going to

01:22:00.489 --> 01:22:05.069
fall this year. What was your last prediction,

01:22:05.390 --> 01:22:08.810
Martin? Oh, this is one. It's a fun one. I'm

01:22:08.810 --> 01:22:10.470
opting out of this as well, Martin. You just

01:22:10.470 --> 01:22:13.329
tell the listeners what you think. So this is

01:22:13.329 --> 01:22:17.649
a bold one as... Someone who is in and around

01:22:17.649 --> 01:22:22.770
the industry. But my prediction is that UK marketing

01:22:22.770 --> 01:22:28.630
agency headcount will fall over the next 12 months.

01:22:28.869 --> 01:22:30.850
There's a difficult one to measure. The best

01:22:30.850 --> 01:22:34.350
metric that I can find is from the Institute

01:22:34.350 --> 01:22:37.770
of Practitioners in Advertising who have an agency

01:22:37.770 --> 01:22:41.800
census. And they have their 2024 figure that

01:22:41.800 --> 01:22:47.699
I found, which was 26 ,787 employees across their

01:22:47.699 --> 01:22:51.579
members. I predict that number will fall. I think

01:22:51.579 --> 01:22:59.079
we will see more companies doing more in -house.

01:22:59.140 --> 01:23:01.840
So they'll make their marketing teams more efficient.

01:23:02.039 --> 01:23:06.640
But I also, that's client side. And I also think

01:23:06.640 --> 01:23:11.199
agencies. will themselves be able to do more

01:23:11.199 --> 01:23:15.479
with less. And as such, through the use of things

01:23:15.479 --> 01:23:19.399
like agentic teams, I think there will be reduced

01:23:19.399 --> 01:23:25.840
demand for headcount. Over to you, Paul. Obviously,

01:23:25.859 --> 01:23:27.420
as someone who works within a marketing agency

01:23:27.420 --> 01:23:31.100
group, I don't have an opinion, but obviously

01:23:31.100 --> 01:23:34.699
I always have an opinion regardless. Do you know

01:23:34.699 --> 01:23:40.029
what I can't decide? is with more capabilities

01:23:40.029 --> 01:23:44.090
to get stuff done, to reach customers better,

01:23:44.250 --> 01:23:48.170
to enter new markets, that as long as demand

01:23:48.170 --> 01:23:50.550
keeps up, and there's a big if on that, honestly,

01:23:50.670 --> 01:23:56.470
but if it does, or if it were to, I could see

01:23:56.470 --> 01:23:59.329
a world where expectations, workloads and output

01:23:59.329 --> 01:24:04.109
scale with the new capabilities. So rather than

01:24:04.109 --> 01:24:06.199
cutting headcount, it's like, cool. you folks

01:24:06.199 --> 01:24:09.079
can now better market to these groups of people,

01:24:09.199 --> 01:24:12.779
help us be more successful in all of these verticals.

01:24:12.779 --> 01:24:13.939
And by the way, we're going to enter some new

01:24:13.939 --> 01:24:16.920
markets because we've, you know, now we can do

01:24:16.920 --> 01:24:18.979
these things at a scale across our entire business,

01:24:19.060 --> 01:24:21.020
to be honest, not just marketing that we couldn't

01:24:21.020 --> 01:24:24.859
do before. I think eventually demand runs out

01:24:24.859 --> 01:24:27.680
when you go down that route, but in the short

01:24:27.680 --> 01:24:29.600
to medium term, especially for early movers,

01:24:29.800 --> 01:24:34.810
I actually think the ability to like 10x the

01:24:34.810 --> 01:24:38.210
impact of your competitors is actually what makes

01:24:38.210 --> 01:24:42.729
you grow. So client side, if you're able to out

01:24:42.729 --> 01:24:44.409
-compete the market through effective marketing

01:24:44.409 --> 01:24:47.149
by using agent teams, I think you'll grow. And

01:24:47.149 --> 01:24:48.850
I think you'll hire people to help you do even

01:24:48.850 --> 01:24:52.470
more of that. The same for agencies. I think

01:24:52.470 --> 01:24:54.850
for agencies that stick their head in the sand

01:24:54.850 --> 01:24:57.170
or for client teams that stick their head in

01:24:57.170 --> 01:25:00.390
the sand, at some point, they'll be asked hard

01:25:00.390 --> 01:25:04.439
questions and... I don't know that, you know,

01:25:04.439 --> 01:25:07.500
businesses will then have to try and adopt AI

01:25:07.500 --> 01:25:09.640
and whether they'll outsource to get that in

01:25:09.640 --> 01:25:11.539
the first instance, which has happened a lot

01:25:11.539 --> 01:25:14.600
over history, which again would suggest agencies

01:25:14.600 --> 01:25:17.000
getting more work, not less, at least in the

01:25:17.000 --> 01:25:20.359
short to medium term. So I don't know. It's difficult

01:25:20.359 --> 01:25:24.220
to predict. But the one thing that's true, I

01:25:24.220 --> 01:25:26.479
think, is that you're not getting replaced by

01:25:26.479 --> 01:25:29.060
AI yet, but you will be replaced by someone who

01:25:29.060 --> 01:25:33.909
uses AI if you don't. that's the kind of second

01:25:33.909 --> 01:25:36.970
part of my prediction. And I wasn't sure which

01:25:36.970 --> 01:25:40.229
one to lead with really, because frankly, that

01:25:40.229 --> 01:25:42.989
prediction that I've made there could very well

01:25:42.989 --> 01:25:45.170
come true and it could have absolutely nothing

01:25:45.170 --> 01:25:49.390
to do with AI and everything to do with a changing

01:25:49.390 --> 01:25:51.510
economic landscape, right? There could be an

01:25:51.510 --> 01:25:54.310
economic crash. And you know what? Across the

01:25:54.310 --> 01:25:56.430
board, headcounts fall in all sorts of industries.

01:25:56.609 --> 01:26:00.130
So yeah, that is a difficult one to pinpoint

01:26:00.130 --> 01:26:04.319
exactly on AI. to that though about um replaced

01:26:04.319 --> 01:26:08.279
by people doing ai and using ai and being ai

01:26:08.279 --> 01:26:11.060
literate and an additional part of my prediction

01:26:11.060 --> 01:26:14.439
with regard to the agency space is that i expect

01:26:14.439 --> 01:26:19.699
we are going to see a significant increase in

01:26:19.699 --> 01:26:24.960
the number of job vacancies or placements whatever

01:26:24.960 --> 01:26:30.520
it may be that are head of agentic or similar

01:26:31.690 --> 01:26:37.510
both client side and agency side. Yeah, I agree

01:26:37.510 --> 01:26:40.529
with that. I mean, we had a dedicated section

01:26:40.529 --> 01:26:42.350
on agents, but I think we've touched on them

01:26:42.350 --> 01:26:43.810
enough throughout the podcast so we don't have

01:26:43.810 --> 01:26:45.890
to subject the poor folks to too much more on

01:26:45.890 --> 01:26:49.550
it. But I do think that 2026 is the year where

01:26:49.550 --> 01:26:52.970
we start to chip away at some of those agentic

01:26:52.970 --> 01:26:56.010
use cases in a sort of manageable fashion like

01:26:56.010 --> 01:26:59.210
we've both talked about and that marketers who...

01:26:59.609 --> 01:27:05.010
get to grips early with using claude code or

01:27:05.010 --> 01:27:08.550
core co -work or manus and or and gemini's equivalent

01:27:08.550 --> 01:27:11.189
that they you know they have their i can't remember

01:27:11.189 --> 01:27:12.869
what it's called but they have their like workflow

01:27:12.869 --> 01:27:17.189
work streamed type solution that's kind of zapier

01:27:17.189 --> 01:27:20.069
meets agentic capabilities i think we'll consider

01:27:20.069 --> 01:27:22.630
continue to see that emerge and i think in the

01:27:22.630 --> 01:27:26.060
first instance As marketers, we'll have to be

01:27:26.060 --> 01:27:28.539
proactive in seeking out those solutions, testing

01:27:28.539 --> 01:27:31.880
them and being early adopters and finding where

01:27:31.880 --> 01:27:35.340
those arbitrages are that we can just like get

01:27:35.340 --> 01:27:37.100
a load done through agents and have a bigger

01:27:37.100 --> 01:27:40.279
impact. And then eventually it will be baked

01:27:40.279 --> 01:27:43.100
into Google Gemini or in ChatGPT or whatever

01:27:43.100 --> 01:27:45.479
in a way that brings it to the masses because

01:27:45.479 --> 01:27:48.859
it's much easier to do. The equivalent of send

01:27:48.859 --> 01:27:52.770
your agent co -worker a message in Slack. type

01:27:52.770 --> 01:27:55.029
stuff but i think in the first instance it won't

01:27:55.029 --> 01:27:58.050
be that it will be people playing around with

01:27:58.050 --> 01:28:01.149
cloud code or cloud co -work to figure out what

01:28:01.149 --> 01:28:04.869
can it reliably do for me in my specific context

01:28:04.869 --> 01:28:08.210
and that people who are able to do that to your

01:28:08.210 --> 01:28:12.409
point will be absolutely in massive demand they

01:28:12.409 --> 01:28:15.670
will be the jobs the job specs that go up and

01:28:15.670 --> 01:28:17.449
those will be the people who are getting hired

01:28:17.449 --> 01:28:20.430
who are people who know how to orchestrate agents

01:28:20.430 --> 01:28:23.210
to get stuff done against the capabilities that

01:28:23.210 --> 01:28:26.310
are happening today and also enough on the pulse

01:28:26.310 --> 01:28:30.470
to evolve as the systems evolve it's not the

01:28:30.470 --> 01:28:33.489
head of human resources is the head of agent

01:28:33.489 --> 01:28:36.449
resources and that's the role that we'll see

01:28:36.449 --> 01:28:39.739
in 2026 Well, thanks for sharing your predictions

01:28:39.739 --> 01:28:41.779
with me, Martin. That was the first time I got

01:28:41.779 --> 01:28:43.800
to hear them as well, dear listener. I enjoyed

01:28:43.800 --> 01:28:46.479
them. I hope you enjoyed them as well. And I

01:28:46.479 --> 01:28:47.939
love hanging out with you on these podcasts,

01:28:48.100 --> 01:28:49.920
Martin, so I shall look forward to speaking to

01:28:49.920 --> 01:28:52.260
you again soon. Can't wait. Speak to you soon.

01:28:53.800 --> 01:28:56.600
Thank you for listening to Artificially Intelligent

01:28:56.600 --> 01:28:59.560
Marketing. To stay on top of the latest trends,

01:28:59.779 --> 01:29:02.760
tips and tools in the world of marketing AI,

01:29:03.100 --> 01:29:06.520
be sure to subscribe. We look forward to seeing

01:29:06.520 --> 01:29:07.699
you again next week.