WEBVTT 00:00:00.000 --> 00:00:03.279 For decades, biology had this this dirty little 00:00:03.279 --> 00:00:05.960 secret. Researchers would look at the human genome, 00:00:06.059 --> 00:00:08.759 three billion letters of code, and they basically 00:00:08.759 --> 00:00:12.060 shrug at most of it. Yeah. They told us that 00:00:12.060 --> 00:00:16.079 98 percent of our DNA was just junk. Just biological 00:00:16.079 --> 00:00:19.320 noise, leftovers from evolution that, you know, 00:00:19.320 --> 00:00:21.440 didn't really do anything. It's honestly wild 00:00:21.440 --> 00:00:22.899 when you think about it. It's like reading a 00:00:22.899 --> 00:00:25.019 book and assuming 98 percent of the pages are 00:00:25.019 --> 00:00:27.039 just typos because you don't understand the language. 00:00:27.120 --> 00:00:29.519 Exactly. But then all of a sudden, the lights 00:00:29.519 --> 00:00:31.399 turned on. We just found the instruction manual 00:00:31.399 --> 00:00:34.359 for that 98%. And at the exact same moment, we're 00:00:34.359 --> 00:00:37.200 decoding the biological noise inside us. We are 00:00:37.200 --> 00:00:40.240 drowning in digital noise outside of us. Andres 00:00:40.240 --> 00:00:42.579 Karpathy just dropped this huge warning about 00:00:42.579 --> 00:00:45.659 a slopocalypse coming later this year. Two frontiers 00:00:45.659 --> 00:00:47.820 colliding at once. One unlocking the code inside, 00:00:47.960 --> 00:00:50.560 the other just drowning us in code outside. It 00:00:50.560 --> 00:00:53.259 is going to be a wild Wednesday. Welcome back 00:00:53.259 --> 00:00:56.259 to The Deep Dive. It is Wednesday, January 28th, 00:00:56.259 --> 00:00:59.060 2026. I hope you're having a thoughtful week. 00:00:59.280 --> 00:01:01.500 I'm your host. And I'm ready to break this all 00:01:01.500 --> 00:01:04.290 down. We have a fascinating stack of reports 00:01:04.290 --> 00:01:06.290 today, and it really feels like we are living 00:01:06.290 --> 00:01:09.170 in a moment of incredibly high contrast. Oh, 00:01:09.170 --> 00:01:11.709 totally. We're seeing the microscopic and the 00:01:11.709 --> 00:01:13.989 macroeconomic shifting at the exact same time. 00:01:14.090 --> 00:01:16.489 It's like the interior architecture of, you know, 00:01:16.489 --> 00:01:18.959 of... Humans and the exterior architecture of 00:01:18.959 --> 00:01:20.719 the Internet are both being rewritten simultaneously. 00:01:21.200 --> 00:01:23.980 Exactly. So for you listening in, here is the 00:01:23.980 --> 00:01:26.640 roadmap for today's deep dive. We are going to 00:01:26.640 --> 00:01:30.379 start with DeepMind's Alpha Genome. They have 00:01:30.379 --> 00:01:32.340 essentially turned the lights on in a very dark 00:01:32.340 --> 00:01:35.060 room of human biology. We need to understand 00:01:35.060 --> 00:01:38.640 why that 98 % matters so much. Then we're pivoting 00:01:38.640 --> 00:01:42.290 to... A classic David and Goliath story. A 30 00:01:42.290 --> 00:01:46.010 -person startup called RCAI just took on Meta 00:01:46.010 --> 00:01:48.549 in China for $20 million. Yeah. Spoiler alert. 00:01:48.609 --> 00:01:50.769 They might have actually won. We will also talk 00:01:50.769 --> 00:01:53.790 about the agentic future. Yahoo, Chrome, Amazon, 00:01:53.950 --> 00:01:56.689 everyone is pivoting to agents and it is costing 00:01:56.689 --> 00:02:00.969 jobs. I mean, 16 ,000 people laid off at Amazon 00:02:00.969 --> 00:02:02.989 alone. Right. We have to talk about the economics 00:02:02.989 --> 00:02:05.469 of that. And finally, we'll hit some rapid fire 00:02:05.469 --> 00:02:07.670 tools, including something people are calling 00:02:07.670 --> 00:02:10.090 the Siri killer and the tools you might need 00:02:10.090 --> 00:02:13.669 to survive this slope apocalypse. Okay, let's 00:02:13.669 --> 00:02:15.930 do it. Let's start with the biology. This one 00:02:15.930 --> 00:02:19.169 really caught my attention because junk DNA is 00:02:19.169 --> 00:02:20.689 a term, you know, most of us learned in high 00:02:20.689 --> 00:02:22.629 school. It was just an accepted fact. It was 00:02:22.629 --> 00:02:25.129 dogma. It was written in the textbooks. For the 00:02:25.129 --> 00:02:27.849 longest time, science focused only on that 2 00:02:27.849 --> 00:02:31.310 % of DNA that codes for proteins. The genes that 00:02:31.310 --> 00:02:33.289 make your eyes blue or your hair curly, that 00:02:33.289 --> 00:02:35.430 kind of stuff. Right. The rest, the non -coding 00:02:35.430 --> 00:02:38.530 DNA was treated like packing peanuts, just there 00:02:38.530 --> 00:02:40.830 to fill space. Which in hindsight, calling it 00:02:40.830 --> 00:02:43.490 junk was pretty arrogant of us. We just assumed 00:02:43.490 --> 00:02:46.069 if it didn't make proteins, the building blocks, 00:02:46.229 --> 00:02:49.439 it didn't matter. But nature is rarely that wasteful. 00:02:49.500 --> 00:02:52.439 No, nature doesn't carry around 98 % dead weight 00:02:52.439 --> 00:02:55.199 for millions of years. But now Google DeepMind 00:02:55.199 --> 00:02:59.400 has released Alpha Genome and this, this feels 00:02:59.400 --> 00:03:01.939 like a watershed moment. This is the missing 00:03:01.939 --> 00:03:05.699 piece. Alpha Genome is an AI system that predicts 00:03:05.699 --> 00:03:08.400 how mutations in that mysterious 98 % affect 00:03:08.400 --> 00:03:11.500 gene expression. Okay, let's pause on gene expression. 00:03:11.699 --> 00:03:13.259 I want to make sure everyone really gets that 00:03:13.259 --> 00:03:16.460 concept. When we say it affects expression, what 00:03:16.460 --> 00:03:19.699 does that practically mean? Okay, think of your 00:03:19.699 --> 00:03:22.960 DNA like a massive house with thousands of light 00:03:22.960 --> 00:03:24.939 bulbs. We've known where the light bulbs were 00:03:24.939 --> 00:03:27.039 for a long time. Those are the protein -coding 00:03:27.039 --> 00:03:30.080 genes. But a house isn't just the bulbs. It's 00:03:30.080 --> 00:03:32.740 the wiring behind the walls. It's the dimmer 00:03:32.740 --> 00:03:35.280 switches. It's the whole fuse box. So the junk 00:03:35.280 --> 00:03:38.439 was actually the wiring. Exactly. The non -coding 00:03:38.439 --> 00:03:40.919 DNA tells the genes when to turn on, where to 00:03:40.919 --> 00:03:44.169 turn on, and how brightly to shine. Alpha Genome 00:03:44.169 --> 00:03:46.370 is essentially mapping that wiring for the first 00:03:46.370 --> 00:03:49.770 time. It tells us if you cut this wire here in 00:03:49.770 --> 00:03:52.250 the non -coding region, the light in the kitchen 00:03:52.250 --> 00:03:54.729 goes out. That is a huge shift in perspective. 00:03:55.090 --> 00:03:57.250 The source material used a really helpful analogy 00:03:57.250 --> 00:04:00.310 here about stacking blocks of discovery. I found 00:04:00.310 --> 00:04:02.330 it really helped me contextualize where Alpha 00:04:02.330 --> 00:04:04.669 Genome sits. Yeah, I love this visual. It shows 00:04:04.669 --> 00:04:07.590 how AI has just systematically conquered biology. 00:04:08.150 --> 00:04:11.259 So first you had Alpha Fold. That solved protein 00:04:11.259 --> 00:04:13.439 structure. That was figuring out the actual shape 00:04:13.439 --> 00:04:16.339 of the legal blocks. Which was huge for drug 00:04:16.339 --> 00:04:18.360 discovery, because if you know the shape, you 00:04:18.360 --> 00:04:21.180 know how to attach things to it. Precisely. Then 00:04:21.180 --> 00:04:23.740 came alpha -mescence. That showed us what happens 00:04:23.740 --> 00:04:26.600 when the blocks break or mutate, but only within 00:04:26.600 --> 00:04:28.800 the coding regions. And now alpha -genome. It's 00:04:28.800 --> 00:04:31.420 the final layer. It combines multiple massive 00:04:31.420 --> 00:04:34.699 data sets on gene regulation to show us the rules 00:04:34.699 --> 00:04:37.730 for the non -coding regions. It can predict how 00:04:37.730 --> 00:04:39.889 a single letter change in the junk can cause 00:04:39.889 --> 00:04:42.810 a gene to malfunction. It is staggering when 00:04:42.810 --> 00:04:44.689 you actually think about it. We are effectively 00:04:44.689 --> 00:04:48.250 turning the human genome into a programmable 00:04:48.250 --> 00:04:51.610 map. We are. But I have to ask, why now? I mean, 00:04:51.629 --> 00:04:53.750 why was this the hardest part to crack? Because 00:04:53.750 --> 00:04:56.209 the signal -to -noise ratio is just terrible. 00:04:56.509 --> 00:04:58.949 In coding regions, there are strict rules. Three 00:04:58.949 --> 00:05:01.069 letters equals one amino acid. It's a clear language. 00:05:01.209 --> 00:05:03.310 In the non -coding regions, it just looks like 00:05:03.310 --> 00:05:06.540 random chaos. You need massive transformer models, 00:05:06.699 --> 00:05:09.399 basically the same tech behind ChatGPT, to look 00:05:09.399 --> 00:05:11.920 at that chaos and find the subtle patterns that 00:05:11.920 --> 00:05:14.600 human eyes could never see. Now, we should temper 00:05:14.600 --> 00:05:16.899 the excitement just a little bit. The report 00:05:16.899 --> 00:05:19.360 mentions this isn't clinical grade yet, right? 00:05:19.459 --> 00:05:21.300 You can't go to your doctor tomorrow and get 00:05:21.300 --> 00:05:23.839 an alpha genome scan. Correct. It is strictly 00:05:23.839 --> 00:05:26.500 a research tool right now. And it's currently 00:05:26.500 --> 00:05:30.439 only trained on humans and mice. So don't expect 00:05:30.439 --> 00:05:33.339 it to cure rare diseases by next Tuesday. But 00:05:33.339 --> 00:05:35.420 that raises a question for me. If it's just a 00:05:35.420 --> 00:05:38.360 research tool for now, why is this a headline 00:05:38.360 --> 00:05:40.560 breakthrough? Why should the person listening 00:05:40.560 --> 00:05:44.079 care about a mouse model tool? Because it unlocks 00:05:44.079 --> 00:05:46.779 the 98 % of instructions that actually control 00:05:46.779 --> 00:05:49.500 our biology. That is the key. It shifts the entire 00:05:49.500 --> 00:05:52.180 map of discovery. It does. Speaking of shifting 00:05:52.180 --> 00:05:54.480 maps and finding efficiency where we didn't think 00:05:54.480 --> 00:05:57.040 it existed, let's pivot to the world of large 00:05:57.040 --> 00:05:59.139 language models. Let's do it. From biological 00:05:59.139 --> 00:06:02.149 code to synthetic code. There has been this narrative 00:06:02.149 --> 00:06:05.209 for a few years now, the scaling laws. You know, 00:06:05.250 --> 00:06:07.269 if you want to build a frontier model, something 00:06:07.269 --> 00:06:10.870 that can reason, code, write poetry, you need 00:06:10.870 --> 00:06:13.250 billions of dollars. You need to be open AI, 00:06:13.509 --> 00:06:16.829 anthropic or meta. You need a warehouse full 00:06:16.829 --> 00:06:20.430 of H -100s and the GDP of a small country. That's 00:06:20.430 --> 00:06:23.790 the rule. Or so we thought. Enter RCAI. This 00:06:23.790 --> 00:06:27.610 story is just... They're a scrappy 30 -person 00:06:27.610 --> 00:06:29.810 startup. They don't have a massive campus or 00:06:29.810 --> 00:06:32.649 Microsoft backing them. And yet they just released 00:06:32.649 --> 00:06:36.089 a model called Trinity. A 400 billion parameter 00:06:36.089 --> 00:06:40.750 model built from scratch in six months. For $20 00:06:40.750 --> 00:06:43.610 million. That number is just absurdly low in 00:06:43.610 --> 00:06:46.310 this industry. Usually $20 million is the coffee 00:06:46.310 --> 00:06:49.509 budget for these training runs. We are used to 00:06:49.509 --> 00:06:52.269 seeing training costs. In the hundreds of millions. 00:06:52.509 --> 00:06:54.310 So how did they do it? Is it some sort of magic 00:06:54.310 --> 00:06:55.990 trick or did they actually find a better way 00:06:55.990 --> 00:06:58.149 to bake the cake? It's efficiency over brute 00:06:58.149 --> 00:07:00.629 force. The CEO, Mark McQuaid, he's an interesting 00:07:00.629 --> 00:07:02.949 character. He came from Hugging Face. He realized 00:07:02.949 --> 00:07:04.769 that most of the data giants feed into these 00:07:04.769 --> 00:07:08.230 models is actually, well, junk. Kind of like 00:07:08.230 --> 00:07:09.870 the junk DNA we were just talking about. So they 00:07:09.870 --> 00:07:13.029 just curated the data better. Aggressively. They 00:07:13.029 --> 00:07:16.540 focused on high quality, dense information. They 00:07:16.540 --> 00:07:20.040 also used really advanced model merging techniques, 00:07:20.360 --> 00:07:22.800 taking the best parts of different training runs 00:07:22.800 --> 00:07:25.180 and stitching them together. And his motivation 00:07:25.180 --> 00:07:27.639 was fascinating, too. It wasn't just about saving 00:07:27.639 --> 00:07:30.500 money. No, it was ideological. He wanted independence. 00:07:31.319 --> 00:07:33.160 He didn't want to rely on China -sourced models. 00:07:33.259 --> 00:07:35.980 And he was tired of Meta changing their licenses 00:07:35.980 --> 00:07:38.540 all the time. He wanted a sovereign U .S. model. 00:07:38.680 --> 00:07:40.759 Yeah, so he just built his own. And the performance 00:07:40.759 --> 00:07:43.600 isn't just good for a startup. No, it's legitimately 00:07:43.600 --> 00:07:46.500 competitive. It benchmarks against Meta's Lama 00:07:46.500 --> 00:07:49.800 4 Maverick. It also goes toe -to -toe with China's 00:07:49.800 --> 00:07:52.800 GLM 4 .5. And the report says it actually wins 00:07:52.800 --> 00:07:56.480 in some key areas. In coding, math, and reasoning 00:07:56.480 --> 00:07:59.240 tasks, you know, the hard stuff, it beats them. 00:07:59.319 --> 00:08:01.660 It might lose on creative poetry or writing a 00:08:01.660 --> 00:08:04.899 generic email, but for logic, it's a beast. That 00:08:04.899 --> 00:08:06.779 is significant. It comes in three flavors, right? 00:08:06.899 --> 00:08:09.170 Yeah. You've got base. which is the raw model, 00:08:09.310 --> 00:08:11.209 large preview, which is instruction tuned to 00:08:11.209 --> 00:08:14.189 be helpful, and then TrueBase. TrueBase. That 00:08:14.189 --> 00:08:15.550 one sounds interesting. What makes that different 00:08:15.550 --> 00:08:18.310 from just a regular base model? It's zero post 00:08:18.310 --> 00:08:21.350 -training. It's designed for labs or companies 00:08:21.350 --> 00:08:23.670 that want a totally clean foundation to build 00:08:23.670 --> 00:08:26.850 on. Okay. See, most models come with safety filters 00:08:26.850 --> 00:08:29.790 or biases baked in by the company that made them. 00:08:30.529 --> 00:08:33.179 TrueBase is pure clay. You can mold it into anything 00:08:33.179 --> 00:08:35.799 you want without fighting the original creator's 00:08:35.799 --> 00:08:37.840 hidden rules. This really disrupts that whole 00:08:37.840 --> 00:08:41.580 bigger is better narrative. It suggests efficiency 00:08:41.580 --> 00:08:43.919 might actually matter more than raw capital. 00:08:44.100 --> 00:08:46.419 It proves you don't need a trillion dollar market 00:08:46.419 --> 00:08:49.159 cap to compete at the frontier. It implies the 00:08:49.159 --> 00:08:52.259 moat that Google and OpenAI built, that we have 00:08:52.259 --> 00:08:55.100 more money than God, moat. might not be so deep 00:08:55.100 --> 00:08:57.179 after all. So let me push you on this. Does a 00:08:57.179 --> 00:08:59.580 $20 million model actually threaten a company 00:08:59.580 --> 00:09:02.539 worth trillions? Or is this just a nice niche 00:09:02.539 --> 00:09:04.899 product? Yes, because it proves efficiency and 00:09:04.899 --> 00:09:07.580 open access can outmaneuver raw cash. It pushes 00:09:07.580 --> 00:09:09.980 the intelligence to the edge. And that efficiency 00:09:09.980 --> 00:09:14.200 is accelerating everything. Which brings us to 00:09:14.200 --> 00:09:17.299 our third segment, the agentic shift. And this 00:09:17.299 --> 00:09:19.669 is where things start to get messy. We are seeing 00:09:19.669 --> 00:09:22.710 a massive rollout of agents. And when we say 00:09:22.710 --> 00:09:24.750 agents, we don't just mean a chatbot you talk 00:09:24.750 --> 00:09:27.970 to. We mean software that goes out and does things 00:09:27.970 --> 00:09:31.470 for you. Right. Action -oriented AI. Yahoo just 00:09:31.470 --> 00:09:35.590 launched Scout, which feels weird to say. I mean, 00:09:35.629 --> 00:09:37.929 Yahoo is back in the innovation game. Which is 00:09:37.929 --> 00:09:39.769 pretty cool, actually. It lets you chat with 00:09:39.769 --> 00:09:42.169 Claude while you search the web. But the key 00:09:42.169 --> 00:09:46.409 thing is... Real links, no hallucinations. It's 00:09:46.409 --> 00:09:48.830 grounded in Yahoo's index. Then there's Google 00:09:48.830 --> 00:09:51.029 Chrome. It's basically becoming an agent now. 00:09:51.190 --> 00:09:53.350 Oh, yeah. Chrome is eating the entire OS. It's 00:09:53.350 --> 00:09:55.809 doing everything. Auto -browse, side panels, 00:09:55.970 --> 00:09:58.110 editing images. It's not just a browser anymore. 00:09:58.210 --> 00:10:00.769 It's a doer. It can log into your Gmail, find 00:10:00.769 --> 00:10:02.750 a receipt, and update a spreadsheet without you 00:10:02.750 --> 00:10:05.409 clicking a button. Exactly. An Airtable -launched 00:10:05.409 --> 00:10:07.870 super agent. that one is for the real power users 00:10:07.870 --> 00:10:11.029 for sure it can run 20 task specific agents at 00:10:11.029 --> 00:10:13.610 once to compile research reports i mean imagine 00:10:13.610 --> 00:10:15.870 having 20 interns working in parallel for 30 00:10:15.870 --> 00:10:18.049 seconds that's super agent and of course mark 00:10:18.049 --> 00:10:21.629 zuckerberg is teasing meta's 2026 plan personal 00:10:21.629 --> 00:10:24.230 super intelligence for commerce which is really 00:10:24.230 --> 00:10:27.070 just code for agents that buy things for you 00:10:27.070 --> 00:10:29.480 and harvest your data They want an agent that 00:10:29.480 --> 00:10:31.279 knows you better than your spouse, so when you 00:10:31.279 --> 00:10:33.879 need toothpaste, it's already ordered. But there 00:10:33.879 --> 00:10:36.159 is a cost to all this speed, and this is where 00:10:36.159 --> 00:10:39.200 the tone of the reports gets a bit darker. Andres 00:10:39.200 --> 00:10:41.700 Karpathy, one of the godfathers of modern AI, 00:10:41.980 --> 00:10:45.559 is warning of a slipocalypse in 2026. I love 00:10:45.559 --> 00:10:48.620 that word, slipocalypse. It's terrifyingly descriptive. 00:10:49.179 --> 00:10:51.580 What exactly is he worried about? It's the supply 00:10:51.580 --> 00:10:54.639 and demand of content. Agents like Claude and 00:10:54.639 --> 00:10:57.559 Codex have just gotten too good too fast. We 00:10:57.559 --> 00:11:00.100 can now generate 1 ,000 blog posts, a million 00:11:00.100 --> 00:11:03.100 comments, or 10 ,000 videos in minutes. We're 00:11:03.100 --> 00:11:05.379 generating content faster than we can possibly 00:11:05.379 --> 00:11:08.580 process it. Exactly. We are flooding the entire 00:11:08.580 --> 00:11:12.299 digital ecosystem with slop. Low -quality AI 00:11:12.299 --> 00:11:14.879 -generated noise designed to be read by other 00:11:14.879 --> 00:11:17.059 algorithms, not humans. It's the dead internet 00:11:17.059 --> 00:11:19.480 theory coming true. Agents writing slop for other 00:11:19.480 --> 00:11:21.779 agents to index. And this sufficiency is hitting 00:11:21.779 --> 00:11:25.039 the workforce hard. Amazon just laid off 16 ,000 00:11:25.039 --> 00:11:27.159 people. That's the second major cut in three 00:11:27.159 --> 00:11:30.100 months? It is. And the internal memo was leaked. 00:11:30.620 --> 00:11:33.799 They explicitly said it was too streamlined for 00:11:33.799 --> 00:11:36.840 the AI race. That is the brutal reality of this 00:11:36.840 --> 00:11:39.519 agentic shift. Companies are trading headcount 00:11:39.519 --> 00:11:41.480 for compute. They look at the budget and say, 00:11:41.620 --> 00:11:45.500 why pay a middle manager 100K when 10K of GPU 00:11:45.500 --> 00:11:48.179 credits can run an agent that does the same reporting? 00:11:48.460 --> 00:11:50.879 It's a massive capital reallocation. And it's 00:11:50.879 --> 00:11:52.679 not just Amazon. I mean, look at Anthropic. They 00:11:52.679 --> 00:11:56.220 just raised $20 billion at a $350 billion valuation. 00:11:56.720 --> 00:11:58.820 But look at their cash flow. They don't expect 00:11:58.820 --> 00:12:01.740 to be positive until 2028. So it's a high -stakes 00:12:01.740 --> 00:12:05.139 gamble, high revenue forecasts, $55 billion for 00:12:05.139 --> 00:12:09.240 2026, but a massive, massive burn. It's a land 00:12:09.240 --> 00:12:12.120 grab. And the territory is our attention and 00:12:12.120 --> 00:12:14.519 our jobs. They are betting that whoever builds 00:12:14.519 --> 00:12:17.460 the best agent controls the whole economy. I 00:12:17.460 --> 00:12:19.220 have to admit, reading about all these super 00:12:19.220 --> 00:12:22.080 agents that can run 20 tasks at once, it makes 00:12:22.080 --> 00:12:24.159 me feel a bit inadequate. Oh, totally. There 00:12:24.159 --> 00:12:25.779 was a piece in the source material about the 00:12:25.779 --> 00:12:28.720 real reason you're still slow with AI. I related 00:12:28.720 --> 00:12:31.740 to that so hard. Even you. You track this stuff 00:12:31.740 --> 00:12:33.279 for a living. You're supposed to be the cyborg 00:12:33.279 --> 00:12:36.899 here. Absolutely. I look at these tools. Airtable 00:12:36.899 --> 00:12:39.980 running 20 agents. And I'm still just trying 00:12:39.980 --> 00:12:42.379 to write a good prompt. I find myself falling 00:12:42.379 --> 00:12:44.879 into old habits. Right. I'll default to doing 00:12:44.879 --> 00:12:47.379 things manually just because it feels safer or 00:12:47.379 --> 00:12:49.059 because I don't quite know how to articulate 00:12:49.059 --> 00:12:51.840 what I want to the AI. It's actually comforting 00:12:51.840 --> 00:12:53.659 to hear that. I think a lot of listeners probably 00:12:53.659 --> 00:12:57.320 feel that same AI anxiety. I still wrestle with 00:12:57.320 --> 00:13:00.100 prompt drift myself. I feel overwhelmed rather 00:13:00.100 --> 00:13:02.799 than optimized half the time. It's the paradox 00:13:02.799 --> 00:13:05.159 of automation, right? The tools get faster, but 00:13:05.159 --> 00:13:06.940 we still have to direct them. And that's the 00:13:06.940 --> 00:13:09.019 bottleneck, the human manager. If you don't know 00:13:09.019 --> 00:13:10.879 what to ask, the speed of the answer doesn't 00:13:10.879 --> 00:13:12.840 really matter. Exactly. Speed without direction 00:13:12.840 --> 00:13:15.559 is just crashing faster. Which leads to a really 00:13:15.559 --> 00:13:18.519 difficult question. If agents do everything... 00:13:18.720 --> 00:13:21.740 The execution, the browsing, the coding. What 00:13:21.740 --> 00:13:23.840 happens to the human workforce mentioned in those 00:13:23.840 --> 00:13:26.799 layoffs? We move from doing the work to managing 00:13:26.799 --> 00:13:29.059 the slop in the strategy. Managing the slop. 00:13:29.179 --> 00:13:33.480 That is a grim but likely very accurate job title 00:13:33.480 --> 00:13:35.659 for the future. It's the new middle management. 00:13:35.820 --> 00:13:37.639 You're not writing the email. You're checking 00:13:37.639 --> 00:13:39.980 the email the bot wrote to make sure it didn't 00:13:39.980 --> 00:13:41.940 hallucinate a meeting with a client that doesn't 00:13:41.940 --> 00:13:44.820 exist. Before we look at the specific tools that 00:13:44.820 --> 00:13:48.320 might help us manage that met... The weapons 00:13:48.320 --> 00:13:50.519 for the slipocalypse. Let's take a quick pause. 00:13:51.299 --> 00:13:53.940 We are back. Let's talk about the toolkit for 00:13:53.940 --> 00:13:57.639 2026. If the internet is becoming this noisy, 00:13:57.799 --> 00:14:00.919 agent -filled jungle, what do we use to cut through 00:14:00.919 --> 00:14:03.840 it all? The survival kit. First up, something 00:14:03.840 --> 00:14:07.580 called Moltbot. Or Claudebot. People are calling 00:14:07.580 --> 00:14:10.440 this the Siri killer. That is a bold claim. We've 00:14:10.440 --> 00:14:12.580 heard Siri killer every year for about a decade. 00:14:12.799 --> 00:14:15.120 We have. But this one is different because it's 00:14:15.120 --> 00:14:17.480 always on. It runs your computer. It sees your 00:14:17.480 --> 00:14:19.940 screen. It remembers your workflows. So it actually 00:14:19.940 --> 00:14:23.379 has context. Deep context. Siri fails because 00:14:23.379 --> 00:14:24.980 it doesn't know what you were doing five minutes 00:14:24.980 --> 00:14:27.500 ago. Multbot knows you just opened that invoice. 00:14:27.620 --> 00:14:30.340 So when you say pay it, it knows what it is. 00:14:30.480 --> 00:14:32.820 That's why it's going viral. Then there is OpenAI's 00:14:32.820 --> 00:14:35.519 Prism. This seems targeted at a very specific 00:14:35.519 --> 00:14:39.340 demographic. This is 100%. For the scientists. 00:14:39.659 --> 00:14:43.440 It's a workspace powered by GPT 5 .2. It allows 00:14:43.440 --> 00:14:46.019 them to write and collaborate on research. Yeah. 00:14:46.059 --> 00:14:48.279 And it's trying to accelerate the kind of discoveries 00:14:48.279 --> 00:14:50.519 we talk about with Alpha Genome. It's designed 00:14:50.519 --> 00:14:53.179 to reduce hallucinations and citations. It connects 00:14:53.179 --> 00:14:56.179 to real academic repositories. It's an attempt 00:14:56.179 --> 00:14:59.100 to keep science grounded in truth while using 00:14:59.100 --> 00:15:03.519 AI for speed. We also have Kimi K2 .5. This one 00:15:03.519 --> 00:15:05.850 is coming out of the Asian markets, right? Yes, 00:15:05.870 --> 00:15:08.830 a new intelligent model. It excels in visual 00:15:08.830 --> 00:15:11.850 understanding and in code. It's really just another 00:15:11.850 --> 00:15:13.970 proof point that the model layer is becoming 00:15:13.970 --> 00:15:16.429 a commodity. Everyone has a smart model now. 00:15:16.570 --> 00:15:19.169 And for the marketers, the people perhaps contributing 00:15:19.169 --> 00:15:22.190 to the Slopocalypse, there's Rumora and Autosend. 00:15:22.649 --> 00:15:25.309 Rumor is fascinating and also kind of terrifying. 00:15:25.529 --> 00:15:28.309 It finds viral videos in your niche and places 00:15:28.309 --> 00:15:30.629 brand comments on them automatically. That sounds 00:15:30.629 --> 00:15:32.769 like it absolutely contributes to the slop, doesn't 00:15:32.769 --> 00:15:35.029 it? That's literally automated noise. It 100 00:15:35.029 --> 00:15:37.509 % contributes to the slop. It's bots talking 00:15:37.509 --> 00:15:40.309 to humans in the comments section. But it's effective. 00:15:40.610 --> 00:15:43.730 And AutoSend handles transactional emails. Boring 00:15:43.730 --> 00:15:45.669 but essential plumbing for the internet. Yeah. 00:15:45.769 --> 00:15:48.330 It ensures the receipt gets to you without a 00:15:48.330 --> 00:15:50.860 human ever having to hit send. When I look at 00:15:50.860 --> 00:15:53.500 this list prism for scientists, automated viral 00:15:53.500 --> 00:15:56.539 comments for marketers, I wonder about access. 00:15:56.899 --> 00:16:00.519 How so? Are these tools for everyone? Or are 00:16:00.519 --> 00:16:03.500 we creating a divide where only the 1 % you see 00:16:03.500 --> 00:16:05.120 in the headlines get the super intelligence? 00:16:05.539 --> 00:16:07.919 They start with the 1%, but agents like Chrome 00:16:07.919 --> 00:16:10.940 make them accessible to everyone. That democratization 00:16:10.940 --> 00:16:13.340 is key. It is. The barrier to entry is dropping 00:16:13.340 --> 00:16:16.220 to zero. So the question isn't, can you access 00:16:16.220 --> 00:16:19.940 it? It's, do you know how to use it? So let's 00:16:19.940 --> 00:16:22.740 zoom out. We have covered a lot of ground today. 00:16:22.860 --> 00:16:25.279 From the microscopic DNA to the macroeconomic 00:16:25.279 --> 00:16:29.340 layoffs. A lot. The theme of 2026 so far really 00:16:29.340 --> 00:16:31.700 seems to be the interior versus the exterior. 00:16:32.019 --> 00:16:33.559 That's a great way to frame it. On the interior, 00:16:33.679 --> 00:16:36.419 you have DeepMind unlocking the dark matter of 00:16:36.419 --> 00:16:39.220 our biology with Alpha Genome. We are finally, 00:16:39.259 --> 00:16:41.960 finally reading the code that actually runs us. 00:16:42.179 --> 00:16:44.159 We're looking inward and finding a programmable 00:16:44.159 --> 00:16:47.679 map. And on the exterior, we have agents. Yahoo, 00:16:47.879 --> 00:16:51.460 Chrome, Moltbot managing our digital lives. And 00:16:51.460 --> 00:16:54.899 potentially leading us right into a slopocalypse 00:16:54.899 --> 00:16:58.379 of infinite noise where we have to fight to keep 00:16:58.379 --> 00:17:01.100 our heads above water. It is this tension between 00:17:01.100 --> 00:17:04.380 understanding our natural code and being completely 00:17:04.380 --> 00:17:07.400 overwhelmed by our synthetic code. And the human 00:17:07.400 --> 00:17:10.980 is stuck right in the middle. Just trying to 00:17:10.980 --> 00:17:12.880 figure out which buttons to push. So what is 00:17:12.880 --> 00:17:14.839 the takeaway for you listening to this? If you're 00:17:14.839 --> 00:17:17.119 feeling that anxiety you mentioned earlier. Don't 00:17:17.119 --> 00:17:19.339 just let the agents run on autopilot. That's 00:17:19.339 --> 00:17:21.460 the trap. Right. Look at those habits that are 00:17:21.460 --> 00:17:24.859 keeping you slow. Be the manager, not the bystander. 00:17:25.180 --> 00:17:27.319 If you just watch, you're going to get buried 00:17:27.319 --> 00:17:29.940 in the slop. You have to be the editor of your 00:17:29.940 --> 00:17:32.200 own digital life. You have to direct the flow. 00:17:32.259 --> 00:17:34.559 You have to maintain your own agency. Exactly. 00:17:34.960 --> 00:17:37.799 Agency over agents. I want to leave you with 00:17:37.799 --> 00:17:40.180 a thought today. We talked about a 30 -person 00:17:40.180 --> 00:17:43.000 startup beating meta. We talked about an AI reading 00:17:43.000 --> 00:17:46.819 the 98 % of DNA we ignored for 50 years. Two 00:17:46.819 --> 00:17:48.480 things that were considered completely impossible 00:17:48.480 --> 00:17:52.440 just a few years ago. Exactly. So if RC can beat 00:17:52.440 --> 00:17:55.599 the giants with efficiency and Alpha Genome can 00:17:55.599 --> 00:17:59.339 read the unreadable, what other impossible barriers 00:17:59.339 --> 00:18:01.180 in your life or your business are just illusions 00:18:01.180 --> 00:18:03.619 waiting to break this year? That is the question. 00:18:03.900 --> 00:18:05.220 Thanks for diving in with us. We'll see you in 00:18:05.220 --> 00:18:05.559 the next one.