WEBVTT 00:00:00.000 --> 00:00:01.960 You've probably heard the term AI agent everywhere 00:00:01.960 --> 00:00:04.700 lately. It's one of those phrases people love 00:00:04.700 --> 00:00:08.220 to drop in articles, LinkedIn posts, and conference 00:00:08.220 --> 00:00:11.140 talks. But if you ask five people what it actually 00:00:11.140 --> 00:00:13.820 means, you'll probably get seven different answers. 00:00:14.419 --> 00:00:18.059 All slightly conflicting. Today I want to cut 00:00:18.059 --> 00:00:20.839 through that noise. I'll explain in plain and 00:00:20.839 --> 00:00:23.620 practical language what an AI agent really is, 00:00:23.719 --> 00:00:28.399 how to recognize one, and more importantly, when 00:00:28.399 --> 00:00:31.160 you should use one. And because I believe in 00:00:31.160 --> 00:00:33.820 showing, not just telling, I'll walk you through 00:00:33.820 --> 00:00:36.380 a live demo of a simple agent that triages my 00:00:36.380 --> 00:00:39.799 calendar and my inbox. This tiny agent buys me 00:00:39.799 --> 00:00:42.259 back two hours a week, and I'll show you how 00:00:42.259 --> 00:00:45.399 to build something similar for yourself. No hype, 00:00:45.560 --> 00:00:48.619 no sci -fi, just enough structure that you can 00:00:48.619 --> 00:00:56.380 create your first agent today. Welcome to Data 00:00:56.380 --> 00:00:58.579 and AI with Mukundan, where you learn AI by building. 00:00:59.159 --> 00:01:02.039 Hey, I'm Mukundan. This show is about solving 00:01:02.039 --> 00:01:05.379 real problems with small, useful AI, things that 00:01:05.379 --> 00:01:08.519 you can actually use today. Each week, we pick 00:01:08.519 --> 00:01:11.579 one problem, build a simple workflow or tool, 00:01:11.719 --> 00:01:14.120 and talk through the decisions, what to automate, 00:01:14.340 --> 00:01:17.659 how to check quality, and how to make it reliable. 00:01:18.680 --> 00:01:22.060 If you're a builder, creator, or just AI curious, 00:01:22.459 --> 00:01:25.200 you'll leave with the steps that you can copy 00:01:25.200 --> 00:01:29.120 tonight. I've spent years working in data roles, 00:01:29.299 --> 00:01:31.659 analytics, experimentation, marketing science. 00:01:32.180 --> 00:01:34.540 And one thing that I've learned is that I don't 00:01:34.540 --> 00:01:36.980 care much for buzzwords. I care about outcomes. 00:01:37.700 --> 00:01:40.560 When the idea of AI agents started hitting my 00:01:40.560 --> 00:01:43.620 feeds, I noticed two very different camps forming. 00:01:44.359 --> 00:01:47.879 The first was the hype team declaring that agents 00:01:47.879 --> 00:01:50.780 will replace every kind of work we do. The second 00:01:50.780 --> 00:01:55.659 was the skeptics. dismissing agents as nothing 00:01:55.659 --> 00:01:58.099 more than a chatbot with a few extra steps as 00:01:58.099 --> 00:02:00.200 with most things the truth lies somewhere between 00:02:00.200 --> 00:02:02.659 agents are useful when they can actually plan 00:02:02.659 --> 00:02:06.319 across multiple steps reliably call tools keep 00:02:06.319 --> 00:02:08.919 track of short -term memory and adjust based 00:02:08.919 --> 00:02:12.039 on feedback if what you're building can't do 00:02:12.039 --> 00:02:14.500 those four things then you don't really have 00:02:14.500 --> 00:02:17.240 an agent you just have a fancy auto complete 00:02:17.240 --> 00:02:19.139 but if you're looking for a definition that you 00:02:19.139 --> 00:02:21.719 can explain to your boss here's the definition 00:02:21.719 --> 00:02:25.819 i use Because it passes the explain it to your 00:02:25.819 --> 00:02:28.439 boss test. An AI agent is a software that takes 00:02:28.439 --> 00:02:30.960 a goal, plans the steps it needs to complete 00:02:30.960 --> 00:02:35.159 that goal, uses tools to take action, remembers 00:02:35.159 --> 00:02:38.560 what has already happened, and adapts using feedback 00:02:38.560 --> 00:02:41.879 until the job is finished. I'll repeat that again. 00:02:42.039 --> 00:02:45.039 An AI agent is a software that takes a goal, 00:02:45.340 --> 00:02:49.219 plans the steps that it needs to complete that 00:02:49.219 --> 00:02:54.139 goal, uses tools to take action, remembers what 00:02:54.139 --> 00:02:57.439 has already happened, and adapts using feedback 00:02:57.439 --> 00:02:59.860 until the job is finished. That distinction matters 00:02:59.860 --> 00:03:05.080 because chatbots reply, but agents execute. Think 00:03:05.080 --> 00:03:10.259 of five essential pillars that make this possible. 00:03:10.800 --> 00:03:14.099 First, you need a clear goal or task, like scheduled 00:03:14.099 --> 00:03:17.580 my podcast prep blog this week. Second, you need 00:03:17.580 --> 00:03:20.039 planning. which means breaking that goal into 00:03:20.039 --> 00:03:24.199 concrete steps, such as check the calendar, group 00:03:24.199 --> 00:03:27.960 tasks, propose time blocks, and send invites. 00:03:28.639 --> 00:03:32.240 Third, you need tools and actions, things like 00:03:32.240 --> 00:03:35.520 calendar APIs, email connectors, or document 00:03:35.520 --> 00:03:41.020 fetches. Fourth, you need memory or state, which 00:03:41.020 --> 00:03:44.699 records what's already been done, what failed, 00:03:44.780 --> 00:03:48.780 and what to try next. Finally, you need feedback 00:03:48.780 --> 00:03:52.759 and control. Critique loops, approval checkpoints, 00:03:52.840 --> 00:03:54.939 and retry logic that keep the system safe and 00:03:54.939 --> 00:03:57.719 useful. Where agents usually fail is when goals 00:03:57.719 --> 00:04:02.080 are vague. That guardrails are missing. Tools 00:04:02.080 --> 00:04:04.939 fail without retries or actions aren't locked. 00:04:05.120 --> 00:04:07.900 If your success criteria are unclear, your agent 00:04:07.900 --> 00:04:11.159 will be unclear too. Now, let's look at what 00:04:11.159 --> 00:04:14.400 is an agent versus automation versus an assistant. 00:04:15.599 --> 00:04:17.560 It helps to draw a line between three things. 00:04:18.879 --> 00:04:22.720 Automation, assistants, and agents. Automation 00:04:22.720 --> 00:04:25.579 is a set of fixed steps that run the same way 00:04:25.579 --> 00:04:28.639 every time, which is perfect when inputs are 00:04:28.639 --> 00:04:31.019 stable and predictable. An assistant like a chatbot 00:04:31.019 --> 00:04:33.860 can hold a flexible conversation, but usually 00:04:33.860 --> 00:04:37.620 cannot handle multi -step execution. An agent, 00:04:37.639 --> 00:04:40.600 on the other hand, is both flexible and action 00:04:40.600 --> 00:04:44.129 capable. It can plan, choose tools and iterate 00:04:44.129 --> 00:04:47.649 until the task is complete a good litmus test 00:04:47.649 --> 00:04:50.649 is this if you remove one of its tool connections 00:04:50.649 --> 00:04:53.709 can it still make progress by planning another 00:04:53.709 --> 00:04:57.370 way well if the answer is yes you probably have 00:04:57.370 --> 00:05:00.990 a real agent the answer is no then what you built 00:05:00.990 --> 00:05:04.029 is really just a brittle macro now let's look 00:05:04.029 --> 00:05:08.910 at tooling landscape without the buzz let's talk 00:05:08.910 --> 00:05:12.009 about the actual tooling stripped of all jargon 00:05:13.290 --> 00:05:16.730 At the heart of many agents are planners or controllers. 00:05:17.629 --> 00:05:20.189 These might be large language models prompting 00:05:20.189 --> 00:05:22.209 themselves in loops or more structured planners 00:05:22.209 --> 00:05:26.889 like trees and graphs that map decisions. Agents 00:05:26.889 --> 00:05:28.829 rely on connectors to reach into your ecosystem. 00:05:29.129 --> 00:05:32.269 That is your email, calendar, Google Drive, Notion, 00:05:32.410 --> 00:05:37.009 Slack, databases or the broader web. For memory 00:05:37.009 --> 00:05:41.259 or... For memory, agents often use a simple scratchpad 00:05:41.259 --> 00:05:44.160 to hold state, a vector database for longer -term 00:05:44.160 --> 00:05:48.319 recall or even task graphs to keep track of dependencies. 00:05:49.319 --> 00:05:52.600 Safety is non -negotiable. That means running 00:05:52.600 --> 00:05:56.660 in sandboxes, using approval steps, limiting 00:05:56.660 --> 00:06:00.639 permissions, setting rate limits, and running 00:06:00.639 --> 00:06:04.339 dry runs before executing live. A practical rule 00:06:04.339 --> 00:06:07.689 is to begin with just one or two tools. tie them 00:06:07.689 --> 00:06:11.569 to a single tight success metric. Well, I mean, 00:06:11.610 --> 00:06:14.329 as most people would say, just if you can't describe 00:06:14.329 --> 00:06:17.550 your journey, if you can't describe your user 00:06:17.550 --> 00:06:21.149 journey in three boxes, your input, agent, and 00:06:21.149 --> 00:06:24.410 output, then your scope is really too big. Let's 00:06:24.410 --> 00:06:28.610 look at some five everyday agent demos. To make 00:06:28.610 --> 00:06:32.170 this real, let's look at five everyday agents 00:06:32.170 --> 00:06:35.230 that you can build. One is a calendar blocker 00:06:35.230 --> 00:06:38.089 agent. which reads your goals, scans your schedule, 00:06:38.310 --> 00:06:40.790 proposes time blocks, and asks for your approval 00:06:40.790 --> 00:06:44.990 before then going ahead and creating those events. 00:06:45.930 --> 00:06:49.629 An email triage event labels messages by intent. 00:06:50.930 --> 00:06:55.189 Draft smart replies and queues them for your 00:06:55.189 --> 00:06:58.170 review. A podcast prep agent gathers guest bios, 00:06:58.449 --> 00:07:01.810 compiles a briefing, and drafts questions so 00:07:01.810 --> 00:07:05.519 that you're never unprepared. An inbox to Notion 00:07:05.519 --> 00:07:08.319 connector automatically converts emails into 00:07:08.319 --> 00:07:11.860 tasks with the due dates so that nothing falls 00:07:11.860 --> 00:07:15.759 through the cracks. Finally, a data digest agent 00:07:15.759 --> 00:07:20.040 which pulls yesterday's key metrics, flags anomalies, 00:07:20.100 --> 00:07:22.720 and drafts a summary for Slack so that your team 00:07:22.720 --> 00:07:26.500 gets signal instead of noise. Each of these starts 00:07:26.500 --> 00:07:29.199 small but compounds into serious time savings. 00:07:31.930 --> 00:07:33.910 lightning round to check our instincts before 00:07:33.910 --> 00:07:36.189 we build the live agent. I'll read a statement 00:07:36.189 --> 00:07:39.550 say true or false out loud or in your head if 00:07:39.550 --> 00:07:42.569 you're on a train and I'll give the answer after 00:07:42.569 --> 00:07:47.269 a short pause. An AI agent is just a chatbot 00:07:47.269 --> 00:07:53.189 that answers questions true or false. If I remove 00:07:53.189 --> 00:07:57.209 one tool and the system can still make progress 00:07:57.209 --> 00:08:02.199 by replanning probably looking at a real agent 00:08:02.199 --> 00:08:07.740 so I'm going to repeat this one if I remove one 00:08:07.740 --> 00:08:09.819 tool and the system can still make progress by 00:08:09.819 --> 00:08:12.680 replanning I'm probably looking at a real agent 00:08:12.680 --> 00:08:17.600 true or false right the third one is clear success 00:08:17.600 --> 00:08:23.579 criteria matter less than good prompts true or 00:08:23.579 --> 00:08:28.899 false so clear success criteria matter less than 00:08:28.899 --> 00:08:34.639 good prompts fourth one I have for you chatbots 00:08:34.639 --> 00:08:40.860 reply agents execute across multiple steps I'll 00:08:40.860 --> 00:08:45.259 repeat it chatbots reply agents execute across 00:08:45.259 --> 00:08:50.379 multiple steps so is that true or false automation 00:08:50.379 --> 00:08:54.200 with fixed steps is best when inputs rarely change 00:08:57.779 --> 00:09:00.759 TRF automation with fixed steps is best when 00:09:00.759 --> 00:09:09.559 inputs rarely change all right this is I'm going 00:09:09.559 --> 00:09:12.320 to do one last one here if a tool disconnect 00:09:12.320 --> 00:09:16.679 breaks everything you probably built a brittle 00:09:16.679 --> 00:09:22.940 macro not an agent right so if a tool disconnect 00:09:22.940 --> 00:09:25.580 breaks everything you probably built a brittle 00:09:25.580 --> 00:09:32.139 macro not an ai agent is that true or false so 00:09:32.139 --> 00:09:34.120 i'm going to just give the answers in a second 00:09:34.120 --> 00:09:40.559 but yeah and here's the answers the first one 00:09:40.559 --> 00:09:44.639 where the question was an ai agent is just a 00:09:44.639 --> 00:09:48.940 chatbot that answers questions the answer is 00:09:48.940 --> 00:09:53.399 false the second one was if i remove one tool 00:09:54.080 --> 00:09:56.740 the system can still make progress by replanning. 00:09:57.259 --> 00:10:01.679 I'm probably looking at a real agent. This one 00:10:01.679 --> 00:10:06.799 was true. The third one was clear success criteria 00:10:06.799 --> 00:10:12.200 matter less than good prompts. This one was false. 00:10:13.460 --> 00:10:16.899 The fourth one was chatbots reply agents execute 00:10:16.899 --> 00:10:25.179 across multiple steps. True. Fifth was automation 00:10:25.179 --> 00:10:27.879 with fixed steps is best when inputs rarely change 00:10:27.879 --> 00:10:35.500 This is also true And finally if a tool disconnect 00:10:35.500 --> 00:10:37.620 breaks everything you probably built a brittle 00:10:37.620 --> 00:10:43.899 macro not an agent This is also true So if you 00:10:43.899 --> 00:10:51.519 scored six on six That's awesome But if you haven't, 00:10:51.519 --> 00:10:54.960 just keep listening and I'll be walking through 00:10:54.960 --> 00:10:57.500 AI agents in more depth in the upcoming episodes. 00:10:58.860 --> 00:11:04.820 So now I wanted to do a more live build of a 00:11:04.820 --> 00:11:08.220 calendar plus email triage agent. Let's build 00:11:08.220 --> 00:11:11.440 one together. The goal here is to reclaim two 00:11:11.440 --> 00:11:14.500 hours by batching deep work and clearing administrative 00:11:14.500 --> 00:11:18.279 noise. Success is defined in two ways. By the 00:11:18.279 --> 00:11:20.700 end of the process, I should have two 90 -minute 00:11:20.700 --> 00:11:24.440 defocus blocks for work and one 45 -minute admin 00:11:24.440 --> 00:11:27.100 block added to my calendar for this week. And 00:11:27.100 --> 00:11:31.039 my inbox triage should leave no more than 15 00:11:31.039 --> 00:11:33.759 messages unclassified for my top five replies. 00:11:35.840 --> 00:11:39.559 The plan looks like this. First, read my goals 00:11:39.559 --> 00:11:43.700 for the week from a text note or notion. Next, 00:11:43.759 --> 00:11:46.889 scan my calendar for open slots. Then propose 00:11:46.889 --> 00:11:49.769 blocks aligned with those goals and ask for my 00:11:49.769 --> 00:11:52.850 approval before creating anything. Once approved, 00:11:53.029 --> 00:11:55.789 create events with clear descriptions and embedded 00:11:55.789 --> 00:11:59.389 checklists. After that, pull my 50 most recent 00:11:59.389 --> 00:12:04.490 emails and cluster them by intent. Reply, archive, 00:12:04.830 --> 00:12:08.610 delegate or read later. Finally, draft replies 00:12:08.610 --> 00:12:11.070 for the most valuable threads and queue them 00:12:11.070 --> 00:12:14.639 for review. Guardrails keep the agent safe. Always 00:12:14.639 --> 00:12:20.059 do a dry run first. Only create items in a draft 00:12:20.059 --> 00:12:23.320 or tentative calendar. And never send emails 00:12:23.320 --> 00:12:26.559 without my explicit yes. If the calendar API 00:12:26.559 --> 00:12:29.559 fails, the agent should fall back by writing 00:12:29.559 --> 00:12:32.179 proposed blocks into a document. The email model's 00:12:32.179 --> 00:12:34.360 confidence is too low. It should stop at labeling 00:12:34.360 --> 00:12:36.539 without drafting. If you want to use agents at 00:12:36.539 --> 00:12:39.539 work, see here's the thing. If your agent doesn't 00:12:39.539 --> 00:12:41.419 buy you time or create leverage, it's just a 00:12:41.419 --> 00:12:44.029 toy. Let's look at... how we want to roll this 00:12:44.029 --> 00:12:46.629 out at work without getting fired. If you want 00:12:46.629 --> 00:12:50.389 to use agents at work, it would also depend on 00:12:50.389 --> 00:12:53.490 how your company is open with using AI agents. 00:12:54.710 --> 00:12:56.649 You can start with your own personal workflow 00:12:56.649 --> 00:12:58.769 so that you don't risk production data on day 00:12:58.769 --> 00:13:02.950 one. Ship a tiny agent with just one tool so 00:13:02.950 --> 00:13:05.750 that the scope is easy to control. And always 00:13:05.750 --> 00:13:08.049 add approval steps so that your teammates trust 00:13:08.049 --> 00:13:11.309 the process. Log every action so that nothing 00:13:11.309 --> 00:13:14.830 feels mysterious. And finally, share a one -page 00:13:14.830 --> 00:13:17.830 document that outlines the goal, the scope, and 00:13:17.830 --> 00:13:22.129 the outcomes. This builds confidence and shows 00:13:22.129 --> 00:13:24.049 that you're solving problems, not creating new 00:13:24.049 --> 00:13:26.450 risks. Now it's your turn. I want you to pick 00:13:26.450 --> 00:13:29.210 one recurring pain point. Maybe it's batching 00:13:29.210 --> 00:13:32.570 your calendar, clearing up your inbox, or prepping 00:13:32.570 --> 00:13:35.809 content. Define a success metric that would prove 00:13:35.809 --> 00:13:38.730 this is working. Write out a simple five -step 00:13:38.730 --> 00:13:43.679 plan. Limit yourself to just... one tool. Then 00:13:43.679 --> 00:13:48.460 ship a version 0 .5 today. Remember, the goal 00:13:48.460 --> 00:13:50.340 is not perfection, it's just proof. We should 00:13:50.340 --> 00:13:54.639 always restrict the problem until it becomes 00:13:54.639 --> 00:13:57.080 impossible not to ship. Well, if this episode 00:13:57.080 --> 00:13:59.000 helped you, share it with a teammate or a friend 00:13:59.000 --> 00:14:01.899 who's drowning in admin work. That's exactly 00:14:01.899 --> 00:14:04.980 who benefits from agents first. And stay tuned 00:14:04.980 --> 00:14:09.059 because this week we'll be going even deeper 00:14:09.059 --> 00:14:11.679 into planning techniques. how to make your agents 00:14:11.679 --> 00:14:14.120 reason step by step without wandering off track 00:14:14.120 --> 00:14:26.080 all right quick writing round i'm also going 00:14:26.080 --> 00:14:30.279 to do a more deeper dry run of my calendar and 00:14:30.279 --> 00:14:34.320 inbox but you will see a link to this in the 00:14:34.320 --> 00:14:37.740 show notes of how i've done it for my calendar 00:14:37.740 --> 00:14:42.159 and inbox then you can also try to replicate 00:14:42.159 --> 00:14:48.259 the same here's the plan that I want to do right 00:14:48.259 --> 00:14:52.519 this was no email sent no events created because 00:14:52.519 --> 00:14:56.720 well safety first but the plan is this 290 minute 00:14:56.720 --> 00:14:59.940 deep work blocks on Tuesday and Thursday from 00:14:59.940 --> 00:15:03.960 930 to 11 a .m. a single admin sweep on Thursday 00:15:05.480 --> 00:15:09.440 From 2 45 to 3 30 and a 30 minute workout on 00:15:09.440 --> 00:15:13.600 Wednesday at 12 30 These slots they avoid my 00:15:13.600 --> 00:15:18.200 2 p .m. Agent check -ins Leave my 4 to 4 50 p 00:15:18.200 --> 00:15:24.139 .m. Focus our intact and won't collide With other 00:15:24.139 --> 00:15:28.820 things I have going for me For inbox triage. 00:15:28.960 --> 00:15:32.059 I the agent will classify the messages into reply 00:15:32.059 --> 00:15:38.169 delegate archive or read later, it will only 00:15:38.169 --> 00:15:41.450 draft replies for the top five reply items. Keep 00:15:41.450 --> 00:15:46.049 every draft under 100 words and include exactly 00:15:46.049 --> 00:15:50.590 one next step. If the confidence drops below 00:15:50.590 --> 00:15:55.590 0 .7 on any email, it will label it read later 00:15:55.590 --> 00:16:00.850 and explain why in a single sentence. All of 00:16:00.850 --> 00:16:03.220 this is logged. and i approve before anything 00:16:03.220 --> 00:16:07.000 would go live that's the dry run spec but if 00:16:07.000 --> 00:16:09.000 you want to follow along i pasted the templates 00:16:09.000 --> 00:16:12.340 in the show notes and calendar proposal approval 00:16:12.340 --> 00:16:15.700 checklist and the inbox coach prompt that you 00:16:15.700 --> 00:16:18.679 can paste into your own chat gpt let's hit it 00:16:29.790 --> 00:16:33.409 Before you go, just wanted to let you know that 00:16:33.409 --> 00:16:37.230 I record with Riverside. I'm an affiliate partner 00:16:37.230 --> 00:16:39.570 of Riverside. They helped me record this episode 00:16:39.570 --> 00:16:43.490 and they have amazing audio quality. So that's 00:16:43.490 --> 00:16:45.269 something I just wanted to recommend if you're 00:16:45.269 --> 00:16:48.570 ever deciding to do your own podcast. This is 00:16:48.570 --> 00:16:51.809 a great platform to record because their editing 00:16:51.809 --> 00:16:55.110 features are amazing. Their AI editing is, I 00:16:55.110 --> 00:16:57.669 think, next level. Helps you remove all awkward 00:16:57.669 --> 00:17:02.820 pauses. gives your sound a strong voice it gives 00:17:02.820 --> 00:17:06.680 gives you a strong voice so i would highly recommend 00:17:06.680 --> 00:17:09.740 riverside as the reason why i chose to be an 00:17:09.740 --> 00:17:11.799 affiliate partner because i actually use the 00:17:11.799 --> 00:17:16.799 product and i love it so link to join their platform 00:17:16.799 --> 00:17:20.019 is in the show notes and again i'm an affiliate 00:17:20.019 --> 00:17:23.700 partner so if you make any purchase i may make 00:17:23.700 --> 00:17:27.500 a small commission from it i also host an rss 00:17:27.500 --> 00:17:31.309 .com Also, I'm an affiliate partner of RSS because 00:17:31.309 --> 00:17:37.230 they have an amazing distribution channel. They 00:17:37.230 --> 00:17:42.150 are able to publish your episodes from RSS into 00:17:42.150 --> 00:17:45.369 Spotify, Apple Podcasts, Amazon Music, Audible, 00:17:45.630 --> 00:17:51.490 Deezer, Pandora, iHeartRadio, and even lets you 00:17:51.490 --> 00:17:54.210 publish your episode with one button into YouTube 00:17:54.210 --> 00:17:58.009 as well. Did I forget to mention that you get 00:17:58.490 --> 00:18:01.630 You can start making money by hosting on rss 00:18:01.630 --> 00:18:04.410 .com even if you have just 10 downloads a month. 00:18:05.049 --> 00:18:07.430 So that's all you need to make money, 10 downloads 00:18:07.430 --> 00:18:09.390 a month. It's a new feature they started and 00:18:09.390 --> 00:18:13.009 it's amazing. So again, link to joinrss .com 00:18:13.009 --> 00:18:15.190 will be in the show notes. I'm also using cyda 00:18:15.190 --> 00:18:20.569 .ai to help with my research. So if you are interested 00:18:20.569 --> 00:18:25.509 in any research for your content or in general 00:18:25.509 --> 00:18:28.019 you want to learn about anything, want to read 00:18:28.019 --> 00:18:31.519 research papers or any kind of thing to improve 00:18:31.519 --> 00:18:34.480 your knowledge, cider .ai is your platform to 00:18:34.480 --> 00:18:37.200 use. So again, I'm an affiliate partner and I'd 00:18:37.200 --> 00:18:41.799 love if you're able to support the show by clicking 00:18:41.799 --> 00:18:49.619 on my affiliate link and joining cider .ai. Hey, 00:18:49.779 --> 00:18:52.160 it's Mukundan. If this episode helped you, two 00:18:52.160 --> 00:18:54.819 tiny favors that make a huge difference. Rate 00:18:54.819 --> 00:18:57.720 the show five stars. On Spotify, you can just 00:18:57.720 --> 00:19:00.220 open the show page and tap the star button. On 00:19:00.220 --> 00:19:02.500 Apple Podcasts, scroll to the bottom of the show 00:19:02.500 --> 00:19:05.240 page and tap five stars. Also, leave a one -to 00:19:05.240 --> 00:19:09.039 -line review on Apple Podcasts. Tell me one takeaway. 00:19:09.359 --> 00:19:11.619 I read every single one and it helps more people 00:19:11.619 --> 00:19:14.180 find the show. That's all for this week. Thank 00:19:14.180 --> 00:19:15.799 you for joining and I will see you in the next 00:19:15.799 --> 00:19:17.940 episode of Data and AI with Mukundan, where you 00:19:17.940 --> 00:19:18.680 learn AI by building.