1:04:40Greg Isenberg · Interview
Stop using Claude. Start using Codex?
A 64-minute masterclass where a Codex true believer converts a Claude Code skeptic, live, on camera.
April 27thThe argument in one line.
An AI-native organization is one where people manage agents, agents read and write to a shared context layer, and the system compounds intelligence over time — and the window to build that moat before competitors do is closing fast.
Read if. Skip if.
READ IF YOU ARE…
- You run or work inside a company that has experimented with AI tools but has not seen compounding organizational leverage yet.
- You are a solo founder or small team owner who wants a concrete, replicable framework for what AI-native actually means in practice.
- You are considering starting an AI consulting or acceleration service and need a framework you can sell and deliver.
- You use Claude Code or similar agentic tools and want to understand why skill chains dramatically outperform single prompts.
- You want to see live demos of AI building a proposal microsite and a functional product prototype rather than just slides about it.
SKIP IF…
- You want a step-by-step technical tutorial rather than a framework-plus-demo masterclass.
- You already operate at the skill-chain and company-brain level and want net-new tactics over foundational framing.
The full version, fast.
An AI-native org runs on three layers: people (strategy, taste, judgment), agents (autonomous execution when given clear goals, skills, tools, and context), and a shared context layer (a structured markdown brain that agents read and write to, updated continuously via a capture-curate-store loop). The episode demos the system live: a three-skill chain generates a personalized client proposal microsite in under three minutes; a five-skill chain builds a functional Spotify feature with a usability test layer in under ten. The startup opportunity: verticalize AI acceleration services by niche industry, function, and company size, starting with high-frequency workflows you can show on a sales call.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:22
01 · Intro and promise
Greg frames the masterclass and introduces Theo Tabah of LCA. Three deliverables: what it means to be AI-native, two live workflow demos, and startup ideas.
01:23 – 06:35
02 · DeepMind story and direction
Theo opens with the Demis Hassabis origin story (chess prodigy to Nobel Prize) and his quote from Google IO: running fast in the wrong direction is worse than standing still.
06:36 – 08:18
03 · What is an AI-native org
The three-bullet definition: people manage agents, agents read and write to the company, the company gets smarter. The flywheel: System to Speed to Signal to Moat.
08:19 – 13:22
04 · People layer
Everyone is a manager now. AI eats the execution middle; humans shift to strategy and judgment at the bookends. Pre-AI vs. with-AI work distribution chart.
13:23 – 19:47
05 · Agents layer and skill chains
Agents are models using tools in a loop. Three tiers of use: chat user, approver, autonomous. Four requirements for autonomy: goal, skills, tools, context. LCA skills library on GitHub shown. Skill chains introduced.
19:48 – 30:45
06 · Proposal skill chain demo
Live three-skill chain fires (proposal, copy QA, final QA). A personalized Spotify proposal microsite appears in about two minutes, pulling personal moments from past meeting transcripts stored in the brain. Slack ping arrives with the live URL.
30:46 – 41:07
07 · Context layer: The Recursive Context Layer
Capture (cron pulls from Slack, email, meetings, Linear), Curate (librarian agent reads, files, triggers), Store (markdown brain), Execute (agents leverage context), Experience (signal flows back). Risk: do not let unreviewed AI output loop back into capture.
41:08 – 50:28
08 · Spotify prototype demo
Theo voice-prompts a Daily Blitz feature into Claude Code. Five-skill chain produces a functional Spotify prototype in under 10 minutes with a live usability test layer. AI synthesizes lessons immediately from one completed response.
50:29 – 51:18
09 · Speed impact comparison
Side-by-side table: AI Curious vs AI Native. Proposals: days to minutes. Functional prototype: weeks to minutes. Feedback synthesis: manual to instant.
51:19 – 54:21
10 · Startup ideas: Verticalized AI Acceleration Services
Three niche vectors: industry, function, company size. 2-Up Prioritization Map (niche to general, low to high frequency). Start with niche, high-frequency workflows and show them on sales calls.
54:22 – 56:44
11 · Close and CTA
Greg offers free consultations from Theo for companies at $10M+ ARR. Theo closes: think through the lens of managing agents and what they need to succeed.
Lines worth screenshotting.
- Just using ChatGPT does not make you an AI-native company any more than having a website makes you a tech company.
- AI eats the execution middle of every job, leaving humans to focus on strategy at the front and judgment at the back.
- Everyone is a manager now: your agents will fail for exactly the same reasons a new hire would fail on day one with no context, no tools, and a fuzzy goal.
- Four things determine whether an agent can run autonomously: a clear goal, the right skills, the right tools, and access to the company context.
- A skill chain is a macro skill where each step calls the next, taking you from AI-assisted to AI-native because autonomous agents need composable playbooks, not one-shot prompts.
- The company brain is just folders of markdown files organized so agents can search, retrieve, and write back to them.
- Hallucination drops sharply when you give agents a QA skill as the final step in a chain that checks outputs against source transcripts.
- Speed is the moat: a proposal that used to take three days takes three minutes, and being first with a personalized response has closed Fortune 2000 deals.
- The context layer lets agents remember the detail you forgot, baking personal moments from a meeting months ago into a proposal the prospect never expected.
- Traces and exhaust from agent work are as valuable as finished outputs because they contain the decisions and lessons that never make it into official docs.
- Running 100 miles an hour in the wrong direction is worse than standing still — speed is only valuable when context tells you which direction to run.
- A functional prototype built with Claude Code in under ten minutes gets you real user reactions faster than writing a PRD and waiting weeks.
- To verticalize AI acceleration services, cut your niche on three vectors: industry, function within that industry, and company size.
- Start with niche, high-frequency workflows you can demo on a sales call before expanding to general or low-frequency, high-ROI tasks.
- The 2-Up Prioritization Map separates workflows by niche-to-general and low-to-high frequency, giving a sequenced roadmap for what to sell and build first.
Four things your agents need before they can run on their own.
WHAT TO LEARN
Agents fail for the same reasons a new hire fails on day one with no context, no tools, and a fuzzy goal, and fixing those four inputs is the entire unlock.
- A clear goal is not just a task description: it includes what success looks like, how to measure it, and when it needs to be done; agents without this will approximate and hallucinate.
- Skills are markdown files, nothing more: a playbook, SOP, or reference document that tells the agent what good output looks like; the more specific, the less correction needed.
- Tools unlock what the agent can actually touch: MCP servers, internal APIs, external search; an agent without the right tools will fail even with a perfect goal and perfect skills.
- Context is the company brain: a structured folder tree of markdown files the agent can search and retrieve, built by continuously ingesting Slack, email, meeting transcripts, and project boards through a capture-curate-store loop.
- Skill chains beat single prompts because each step is scoped: a proposal chain that fires proposal generation, then copy QA, then final QA hallucinates far less than one prompt trying to do all three at once.
- The context layer requires a human gate before agent-generated output feeds back into the brain, otherwise the system trains on its own mistakes and drifts.
- Speed becomes a moat only when direction is right: a proposal generated in three minutes from rich meeting context closes deals that a three-day manual proposal would have lost to a competitor who responded first.
- Traces and exhaust, the intermediate artifacts from agent work, are as valuable as finished outputs and should be captured back into the brain as institutional lessons rather than discarded.
- Vertical AI acceleration services have the clearest path to revenue right now: pick an industry, a function within it, and a company size, then build and demo high-frequency workflows on sales calls before expanding scope.
- The 2-Up Prioritization Map sequences what to build first: niche plus high-frequency workflows go first because you can show them on a sales call; general plus high-frequency next; then niche plus low-frequency for its higher ROI.
Terms worth knowing.
- AI-native org
- An organization where people manage agents, agents read and write to the company context, and the combined system gets smarter over time through a continuous capture-curate-store loop.
- Skill
- A markdown file that gives an AI agent a specific capability: a playbook, SOP, or reference document that defines what good output looks like and how to produce it.
- Skill chain
- A macro skill that fires multiple individual skills sequentially, with each skill calling the next, so complex tasks are broken into scoped, quality-controlled steps that dramatically reduce hallucination.
- Company Brain
- A structured folder tree of markdown files representing everything an organization knows: SOPs, meeting notes, client context, lessons, organized so agents can search, retrieve, and write back to it.
- Recursive Context Layer
- The five-stage loop (Capture, Curate, Store, Execute, Experience) that continuously pulls organizational signals into the brain, curates them, makes them agent-readable, and feeds market signal back in.
- Traces / Exhaust
- The intermediate artifacts produced during agent work: explorations, decisions, drafts, that are typically discarded but contain valuable institutional lessons worth storing in the brain.
- Agent autonomy
- The state where an agent runs for extended periods without requiring human approval at each step, achievable only when the agent has a clear goal, the right skills, the right tools, and sufficient context.
- Eval
- A visibility layer into an agent output that compares what was produced against a desired output or quality bar, letting the human-in-the-loop judge whether the work meets the standard.
- Labs page
- A lightweight staging environment where agent-built prototypes are deployed immediately so real users can interact with and test them in the same session the agent builds them.
- Verticalized AI acceleration services
- A service business where an operator implements the AI-native system framework for a specific industry, function, and company size rather than selling general AI consulting.
Things they pointed at.
44:54toolMobin (design system library with MCP)
Lines you could clip.
06:26
“Running 100 miles an hour in the wrong direction is worse than standing still.”
12:03
“Just using ChatGPT does not make you an AI-native company or an AI-native person. That's like having a website and calling yourself a tech company.”
11:11
“Everyone is a manager now.”
23:29
“AI loves to fake it till they make it.”
56:21
“Think through the lens of managing agents and what those agents need to succeed, and you will be well on your way to being ahead of most companies in the world.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00How do you become AI native? In this episode, it is an under sixty minute master class for how to become AI native. This is the type of content that people charge tens of thousands of dollars for, but on this channel, we're giving away for free.
00:16And we're giving it away for free because I believe that people who understand how to become AI native are gonna be able to outperform 99.9% of people on the planet. These are the people that are gonna get raises in an economy where job losses prevail in.
00:30These are the people that are gonna actually create the one person, $1,000,000,000 companies. So in this episode, we break it down in the most clear way possible. You'll learn everything you need to know about how to become AI native.
00:45What does a skill chain mean? What our skills mean? How do I think about context?
00:48How do I pipe things into Claude? And how it all works together. I brought on my cofounder, Theo Taba.
00:54And Theo Taba leads the world around advising the best companies on the planet to become AI native. And in this episode, he spills the sauce.
01:04This is the stuff that, you know, he keeps for ourselves and our team, But I I begged him to come on. He came on, and he does an absolute master class for how to become AI native and explains it in the most clear way I've ever seen on the Internet. So enjoy the episode, and I can't wait to see what you build.
01:23I beg Theo Taba to come on because I think that there's such a huge opportunity in in becoming AI native.
01:39And everyone's saying this word, AI native this, AI native that, but how do you actually become AI native? So Theo is my number one call with this sort of stuff.
01:49So welcome Theo to the pod. By the end of this episode, what are people gonna get out of it?
01:55Nice to see you, Greg. They're gonna get three things. One, how to become an AI native org.
02:00We all wanna know. We're gonna talk about it today. Number two are two workflows in action of how to turn this AI native system that we're gonna talk through into speed that unlock signal from your customers, which is what this is all about.
02:14And, Greg, you're gonna be our signal in this demo and these workflows. And then number three, we gotta talk startup ideas. Right?
02:21So let's talk about a few service based startup ideas. I think, you know, I I'm actually gonna go out on a limb and say this is one of the hottest and fastest growing markets in our lifetime in terms of this space for startup ideas. So I don't wanna come off as, you know, hyperbolic or overblowing anything, but I do think this is a huge opportunity for folks.
02:40So by the end of the episode, people are gonna find out what does it mean to be AI native. What does that concretely mean? And they're also gonna be able to figure out, okay, if I'm a if I wanna be the one person, $1,000,000,000 company, or I'm working in a company, how can I turn that organization into AI native?
02:59And then for the first time ever, you're actually gonna allow us to peek inside of some of the workflows that you are doing within the team that, you know, are things that used to cost millions of dollars to do that you're doing by just becoming AI native, and you're gonna not hold back.
03:19You're gonna share all the sauce. Give me an example of some of the workflows you're gonna show just to give people the taste, and then let's get right into it.
03:27Sure. So here's a prototype. You see this looks a lot like Spotify because that was the system we used.
03:32This prototype, cool feature idea. You can come and listen to music live with your friends.
03:38This is just a demo, of course. This is just one workflow that we're gonna show how to build this in minutes.
03:45This is fully functional, coded up, and then also have it in a full testing suite so you can get direct signal from customers. This is just one of a lot of the things we're gonna talk about today. So that's one kind of hint of what we're gonna be showing.
03:57And and the trick there is it's because you you run an AI native nororg that you were able to get such a high fidelity beautiful prototype, and we'll get into that later. Alright.
04:09Let's go. Let's go. If you'll indulge me for a minute and everyone else, I wanna just tell a very, very quick story that ends with our dear friend Greg here.
04:18So rewind. We're going back to the seventies.
04:21There's a four year old kid, starts playing chess in North London, becomes a master at age 13. Absolutely crushing it. Uses his chess winnings to buy a Commodore Amiga, which is an old computer, new at the time.
04:34Teaches himself to code, builds this amazing game. He's a lead developer on this amazing game called I believe it's called Theme Park.
04:42They sell over a million copies. Makes money from that. Goes to school for computer science.
04:47Oh, the the, you know, person is becoming a little more apparent here, but goes to school for computer science, gets a job, runs a studio, building AI native games, goes back to school, does his PhD in cognitive neuroscience, is fascinated with the brain, wants to know how it works, starts a company, gets funded by none other than Peter Thiel, and they do some incredible stuff, man.
05:08They do some really incredible stuff. They fold every protein on the planet. They create an AI that beats the world's best Go player, which I think has an insane amount of combinations, more atoms than more combinations than atoms in the universe, something like that.
05:24And I think Google then buys them in 2014. This, to me, I heard, was why Elon started OpenAI because he felt like this amount of power in one company's hands was a little too dangerous.
05:41Fast forward to 2024, the dude gets knighted and wins a Nobel Prize. So underachiever. You know?
05:45Real underachiever. Thanks, buddy. Do you know who this person is, Greg?
05:49Well, I know because look at him. And I I was just with him. That's Demis.
05:53You were just with him. Amazing. Yes.
05:55You were just with him. So this is you and Demis at Google IO.
05:59Uh, this is Demis Asabas, cofounder of DeepMind. And who's this, Greg?
06:05It was JD on the Google team. I love that guy. Our boy.
06:09Yeah. JD Choi. Shut up.
06:11Demis had a killer quote at Google IO. Running a 100 miles an hour in the wrong direction is worse than standing still. He did emphasize the importance of speed, but that, again, direction is super important.
06:22And that, I think, ties back to an AI native org and what this is all about. So you can't just run fast for the sake of running fast. You can't just have speed for the sake of speed.
06:31You have to do it in service of your customer, you have to know what you're running to. And this is the magic when you can work so quickly, understand the signal, understand what what you're working towards, and have this AI native system set up, you can deliver incredible value for your customers and, quite frankly, build a moat that makes you unstoppable.
06:53So this is what I've broken down. An AI native org is one where people manage agents. Agents can read and write to the company, and the company gets smarter over time.
07:03That's those are the three bullets. There's, of course, a lot of detail buried in these that we're gonna talk about, but this is the system that allows companies to move with speed and get signal from the market and that creates their moat.
07:15And just because you use ChatGPT does not make you an AI native company or an AI native person. Right?
07:22That's like the the thing you know, I speak to people and I'm and they they say they're AI native, and then I I look into their workflows and it's they're just using ChatGPT.
07:35I I couldn't agree more. It's like if you had a website and called yourself, like, a tech company.
07:41It's just the the gap is massive. So you've good. It's good that you're using that stuff.
07:46Don't get me wrong. But we want to really build this moat.
07:51That's what an AI native org can actually unlock is this system, which is comprised of people, agents, and context. We'll get into each one of those. That produces incredible speed where you can produce anything in minutes.
08:03I flash a little teaser of that. We're gonna do a couple of those. And then signal, we actually get to hear from the market often really quickly, and then all of that feeds back into the system, and this gets smarter and better over time.
08:16Tracking so far? Yes, sir. Amazing.
08:19Okay. Well, let's get into the system. Let's break this down a little bit.
08:23So you had this in your newsletter. I love this.
08:26And it mapped really well, of course, to what an AI native org is comprised of. So you have people at the top. I'm very bullish on people.
08:33I'll get into that in a second. For strategy, taste, judgment, and, of course, that trust piece, you have agents interfacing with the context on behalf of those people.
08:42And that context is really key. I think you called out you have to make your company readable to agents, like AI readable or agent readable.
08:50A lot of people use different terms, consumable, legible, etcetera, etcetera, but I like readable. Let's just stick with that. And this is that shared context layer where the agents essentially have perfect vision, twenty twenty vision on what the company is comprised of.
09:04And that interface between you and that data becomes an incredible level up, and that really allows you to move with speed and, again, get that signal to deliver for customers. All tracking still?
09:17Let's talk. Okay. So let's talk about people.
09:20There's no AI native org without AI native people. Let's just be super clear. Like, obvious, but I think people jump right to, like, let's get the agents in the system.
09:29If your people aren't using this and they're not understanding or do not understand how to use agents and how to use the system, it doesn't matter.
09:39You can put all the tech you want in your company. You can put all the agents, all the AI, all the tools. It's not gonna matter.
09:45And so the big reframe here is what the role of a person becomes. So we have this high level how things were pre AI, where a lot of your work was done in the middle on the execution part.
09:59And a little of it was spent on either side, figuring out what to do, the strategy, etcetera. And then on the end, reviewing the work, is it good enough, is it not good enough, what needs to change, who needs to see it, communicating that work.
10:11The funny thing, though, is, like, these bookends are actually really important. That some might say is the work. That is, like, the really important meaty part of the work.
10:20The research and the drive, all of that, of course, we had to do, but the bookends are really important. So we have this thing where AI actually eats the middle.
10:27And now with AI, you're freed up to focus on the beginning and the end while AI, quote, unquote, eats the middle.
10:37It does all of that execution work on your behalf, And you get to focus on executing and deploying your judgment, your taste, all of that accrued knowledge, and all of the things that make you great as a professional at the beginning and at the end of the work.
10:53Dream come true. Yeah. I think a lot of people
10:57know this now. You know? I think a lot of people are like, okay.
11:00Yes. I feel I understand that my new job is to manage agents, but they're not sure what that really means.
11:07They're not. And I think that's a great point, and that's what we're gonna get into with agents. But I think the main takeaway here is that everyone is a manager now.
11:16And that reframe of making sure your agents are set up for success like a manager would with their team is the unlock in terms of how you look at this. It's not, I've got a new tool like Salesforce or I've got a new tool like Excel.
11:29It's very different than that because, essentially, you have unlimited employees at your disposal, and you need to make sure they're set up for success. Because I think Andy Grove, you know, godfather of management once said, you know, the success of a manager is the success of their team or, you know, judge based to be judge based on the output of their team.
11:47And to me, that is that is it. Or as Greg Eisenberg once said to to my wife,
11:53I really like turndown services at hotels because a turndown service is, like, when they clean your room right before bed, and they basically I have a lot of trouble sleeping.
12:06So, you know, the fact that, like, everything is optimized for the sleep, you know, and, like, everything is, like, perfect.
12:17Never met a turndown service I didn't like. So you wanna you you know?
12:22And I find I sleep better like that. So
12:26I have to bring it back to Seinfeld very quickly. Are you a sheets tucked in or untucked in a hotel room, Greg?
12:33In a hotel room? Yep.
12:38That's a great question. That just feels like a personal question. It is personal.
12:43I'll I'll share I'll share it with, you know, the thousands listening right now. I would say I'm an unpacked person.
12:54I I I don't it's like, I don't need the constraint in my life. Like, don't constrain me. You know?
13:00If I if if especially, I'm six foot three. I know, you know, I know people watch me on YouTube, they're like, when they meet me in real life, they're like, woah, you're large, you know, you're tall. They don't say large.
13:10They say tall. Come on. Give your Yeah.
13:12Exactly. So
13:14I would say untucked, and, yeah, let's keep going.
13:22I'm the same. Alright. Let's talk about agents, that second layer.
13:26You have done, I think, probably the most comprehensive job on the Internet, and I I'm not being you know, you know, we work together.
13:37I think, you know, you know, I'm shooting straight. Compared to everyone on breaking this down, Ross, Mike, Remy, you've had some killer folks on who explain agents, so I'm not gonna spend too much time here. I'm just gonna do a quick refresher and then talk about why they're important.
13:52Agents or models using tools in loop. This is from Barry Zhang at Anthropic. Great engineer.
13:57You gotta give them an environment. You gotta give them tools, and you gotta give them goals. And coming back to everyone's a manager now and how to think about that, I think this kind of overview is really what I wanted to focus on.
14:09So if we look at this, you want your agents right now, there's probably three levels.
14:16You're just chatting with ChatGPT. That's kinda base level or Claude or whomever.
14:22Number two is you've actually got some agents running, and you're sitting there clicking, waiting for the next question to pop up or permission to be granted or prompt or check-in to happen in your cloud code or in your codex. Approve.
14:36Approve. Approve. Maybe you have auto edits on, but someone's just there waiting for an agent to ask you if the next step is okay.
14:43The next state is the agent autonomy, and think about it like a new hire. Right? At the beginning, you're having to babysit them a little bit, giving them what they need.
14:51And then over time, they're actually coming to you with stuff, and they're running for days without your oversight. Maybe weeks, and it's incredible because they're doing great work. And they understand everything, and they're they're absolutely nailing it.
15:03This is what you want your agents to get to. And in order for an agent to have autonomy, they need these four things.
15:09They really need these four things. They need a clear goal.
15:13They need the skills. They need the tools, and they need the context. All of those to succeed and be autonomous.
15:19Again, I will bring it back to your first day on the job. If I walked in to a new company and was expected to put a board deck together for the following week on day one with no management, what would I do?
15:33Maybe the goal is somewhat clear, but a little bit fuzzy. Do I have the skills to do that? Maybe from a past job, but not so much.
15:40Do I have the tools? I don't even know where to start. It's my first day.
15:43Do I have the context? I don't know what's going on with this company I just started. I will fail at that job.
15:48And I think people get impatient with the models or AI because they don't get what they want right away with a very simple prompt or none of this baked in. And so this is really what I wanna harp on.
15:59Again, I know you guys have covered some of this, but I think having all of this baked in, the right goals, the right skills for your agent, the right tools for them, and that context, which we're gonna talk about next, is what unlocks agent autonomy.
16:12The the the other piece on that is, you know, you and and this can go into context, but you don't know what good looks like. So the concept of an eval, can you expand on what that is and why it's an important piece of this whole puzzle?
16:30So an eval is essentially your visibility into the output of an agent. So what did the agent do, and can you see how they got there and what was produced? And
16:40And and the thing that is produced, does how does it like, what is the evaluation of it in terms of, like, is an eight on 10, a nine on 10, a 10 on 10, and comparing it to a desired output.
16:51So that will come from your skills Yeah. The goal Yeah.
16:56And, of course, the context altogether with the right tools. So Yeah.
17:00What I mean by that is if you have a standard, a quality bar, an SOP, this is what good looks like, that can get folded into a skill.
17:09It can also get found in context. Right? It can be a reference document of something that is the pinnacle.
17:15And the goal clearly defines what success looks like when something is great, how to measure if it's great, and when it needs to be great, and when it needs to be done. And so when you combine these things together and then, of course, give the agent the right tools, you actually get the output that you want to the degree or quality you want over and over and over again.
17:36So we have a skills library because, again, this isn't all single player. Right?
17:41When you're an AI native org, you have to think about how the team will benefit from this. How can other folks use agents, and how can those agents use skills? So LCA has a skills library for our org.
17:52This is a demo ish version, meaning it's not fully complete. We didn't wanna show everything. But this is ours, as you know, our our skills library.
18:00So we have a bunch of skills here that people can come in, learn about, and get started. You already know what skills are, so I'm not gonna go too deep into this.
18:08But still our favorite reference is, you know, Neo and the Matrix when they upload kung fu or, you know, combat training directly into his brain. Thank god none of us have the, you know, wires in the back of our necks, but this is essentially what you're doing with agents for skills. Neil loves it.
18:26I love this movie, by the way. So we have a bunch of skills here. And inside, you'll see something that has five skills together.
18:33That's an interesting skill, and that's what we call a skill chain. And, again, skill chains aren't something brand new, but essentially allows you to fire a lot of skills sequentially to make sure that your output is even better. So you don't always want You're gonna cover that later.
18:46Right? We are. We're gonna we're about to get into a demo, man.
18:49We're about to jump right in, and then I'll show you how the skill chains fire. Okay. Cool.
18:53So Yeah. Because I think skill chains
18:56is a really important concept that actually a lot of people haven't covered.
19:01So I'm excited for that. Yeah. As the agents get more autonomous and as the skills and the models get better And as skills can start to call another skill, you can start to have that agent autonomy really start to show up and play a a huge role in how you do the work.
19:18And that's the difference again between that AI native org versus the one that's maybe more AI assisted or AI curious or aren't AI at all. You're just waiting there, and, essentially, you're managing on hard mode.
19:32You're assuming every agent is, an ultra junior. Super smart, but you can't unlock that intelligence, and you're just constantly there trying to direct it, trying to steer it, and it actually just gets frustrating in the end, and maybe you abandon it instead of really having that autonomy. SkillChains allow you to have more autonomous agents.
19:49You've already covered skills and what they are. They're markdown files. You guys know this.
19:53And then skill chains, like I said, are running playbooks back to back. Essentially, it's a macro skill with skills inside of it.
19:59So skill one then fires, call skill two, skill two fires, and then call skill three, skill three fire fires, and then off we go. So I'm gonna give you a demo and a workflow of one thing that we use. Now this is a first SIP workflow.
20:14So, normally, I wouldn't have to touch anything for this to fire. This fires automatically on a trigger.
20:21However, I didn't wanna leave it to a trigger picking something up by chance in this hour. So we're gonna fire it ourselves, and we're gonna just call something in this fake environment that, you know, we at LCA, we work with clients.
20:38And what we're gonna do is pretend there's a new prospect out there. People have heard of Spotify.
20:43Let's just say Spotify is a new prospect. So we're talking to them. We've spoken to them over months, but we haven't actually closed the deal yet.
20:51And now they're ready to get a proposal from us. How are we gonna work together? We've had meetings.
20:56We've spoken about it in Slack. We have figured things out that we need to get done. And, normally, this would fire only on the request for a proposal picked up in a meeting transcript or sent in my inbox.
21:10So it would scan, I'll get into that in the brain and the context very soon. It would pick up that trigger and then fire this skill chain that we were just talking about. So it fires three skills here, and I'm just kind of kicking it off manually.
21:24And in about three minutes, four minutes, we'll actually see the output of this proposal, and I'll break down the skill chain that went into it. But let me jump ahead just to talk to you about that skill chain.
21:37So this proposal flow will get into the capture in a moment when I talk about the brain and the curate. But in the execution phase, which I just triggered, it fires three skills.
21:48Creates a proposal microsite. So, you know, used to send proposals, emails, raw text.
21:54Sometimes that works, but not always. You might want something a little more elevated. Creates a beautiful microsite.
22:00Number two is a copy skill. So make sure that it sounds really tight.
22:05It doesn't sound like AI. It doesn't sound like someone else. It sounds like me and in the conversations we've had.
22:10And number three, a QA skill. So reviews it all, make sure we're not overpromising anything, make sure we're not saying something completely egregious, and make sure we're not making anything up that is not pulled directly from transcripts or from the data.
22:24And so once it's done, it deploys it live on a link, and I can see it. And then it pings me in Slack.
22:32I'll pause there. Do you have any questions on this skill chain before I jump in and see how we're doing with Claude?
22:41I think just the whole concept of a skill chain is, like, people stop at maybe a skill. Right?
22:48Right. And they're missing the chain to actually get high quality stuff. I also think that, you know, a big reason why people, you know, stop using AI as a part of, like, their workflow is they say, well, it hallucinates.
23:05It hallucinates. And this kinda combats a lot of that, I would say.
23:11Totally right. It does. When people say AI hallucinates, one imagine, again, like, an eager new hire who wants to impress you, um, and will just kind of do things to get the job done without considering that it might break break trust.
23:26It's literally fake it until you make it. Right? Exactly.
23:28Like, but exacerbated, like, times a thousand. Exactly.
23:33Yeah. So, yeah, AI loves to fake it till they make it. And Yeah.
23:36Your your job is to make sure that they don't fake it or you minimize that as much as possible.
23:42And I also wanna say one more thing about this proposal thing is LCA, you know, we don't talk about it very publicly, but LCA is I mean, works with literally most of the biggest companies on the planet building AI products, designing, engineering it, and also building AI native orgs.
24:03And a big piece of, like, why we're able to close Fortune two thousands, you know, so frequently is this.
24:13Like, be you know, your LCA is competing against companies that aren't doing this, right, that aren't creating these, like, personalized proposals, going the extra mile.
24:27And, you know, this has been the result of this has been, you know, millions of dollars of of revenue, um, because of this.
24:37So this is, a big deal.
24:39It is. And another thing I just wanna add to that is you and many folks who are, you know, coming from a sales background or sales org knows how important speed is to closing the deal or striking while the iron is hot.
24:54And this is critical. Right? So what normally happens here, what could happen here for companies is someone says, I'd like a proposal.
25:02You have to then go back, review all the notes in between meetings when you have the time. You have to get back to them, say you're on it, you'll get them something, then you have to confirm with the team when there's availability. You have to talk through it.
25:13It might be days before you get them that proposal. They might have cooled off or gone somewhere else. That's just the reality of sales and the reality of the market that we're in and the AI era that we're in.
25:22You can see this already created this. I will risk opening Slack, and it is here. You can see at 10:37AM.
25:30So what is that? A minute or two minutes ago? I got a little note from this is just something I set up.
25:36Aziz is my middle name, and this is, like, a cool I don't I don't wanna get too deep into the story here, but this is my more future guy. And he pings me every time there's a new proposal ready for me to review.
25:48So I'll click on this, and here we go. I'm not saying this is absolutely perfect, but this is the pass that I want. This is the speed to the signal for me.
25:56Is this something that I like and is something that we want? So home and discovery sprint for new listener retention. Boom.
26:02This is, again, a demo. This isn't a real proposal. Spotify has not come to us to ask for this.
26:06I just wanna clear that up. But this is the proposal that gets created. So you've got the outline.
26:12You've got the opportunity.
26:15You've got the whole So,
26:17like, if you sorry. If you scroll if you scroll up, it's like, here's what we're gonna do. We're gonna embed some of our you know, here's the opportunity.
26:25So you you know
26:27Yeah. I can break this down. I'm gonna break it down in a second because I think there's some really cool pieces here.
26:32I wanna give an overview. It's a whole plan week by week, what we're gonna do, the team, how we're gonna do the work, and a little bit about LCA and why us, the cost. You know, I made sure that in the scale, we weren't gonna show real numbers, so we just want Spotify Premium for the team.
26:47And then a little outro. What's the cheapest sprint of all time? What's cool is so I'll give you a few things.
26:54One, it looks pretty good. Like, it's pretty well organized. It's pretty dialed.
26:57It's in the Spotify branding mixed with LCA's branding. So I like that. The spacing, the hierarchy, it all looks pretty dialed, which I love.
27:06So what I like here, I'll bring your attention to it should I asked it to make sure that we bake in some context from the past calls that we've had with Spotify.
27:19So you'll see a line here. So a home that works feels like a record store clerk who knows you. Again.
27:27And the one hands you a record says trust me. So why is this line important? I'm gonna show you something as a preview to the brain or the context section that we're about to cover right after this.
27:37This is a I spun up a brain, put it on GitHub for for this episode. We have a shared LCA brain, but this is one that I I put here. So you can see a brain is just or context is just a bunch of folders with markdown files in them.
27:49Bunch of folders to help guide the agents, read mes. You're essentially guiding the agent through your tree tree structure of folders and files and then helping them land at the right information.
28:03There's a bunch of different search ways or ways to to architect search to go about this, but this is how we've done it. And you can see here in Spotify, you can see correspondence, and you can see things like meetings.
28:16And you can see this one meeting intro call to Maya. Again, this is stuff that we put here to make sure that the proposal could pull from something. So this is the first ever conversation between me and Maya that happened months and months ago.
28:28I learned a little bit about her that call. She's a vinyl person. But here, you can see she said the thing about records or the person behind the counter hands you something and says, trust me.
28:37That's discovery. That's the feeling. Mhmm.
28:39So this is a cool line that she said to me in a meeting that I probably would never really remember when I'm crafting the proposal later on and coordinating with the team.
28:49What's good about this omniscient AI who sees everything has the perfect context is it whips up this proposal and bakes in those little moments of connection that I would love to do given more time.
29:02We would love to do. I I do wanna give this level of personalization, and then you can see it here in these moments. So there's a few of those peppered throughout this proposal, I would hope, because that's what it should do.
29:14It's a direct ask in one of our skills to make sure that it pulls from the transcripts and layers in personalization. You can see here, and good luck in November, save something for mile eight. This is because we know I think there's something in here.
29:29So I run training NYC marathon in November. You know? Again, like and then I think somewhere else, she mentions mile eight and how that's, like, the bet the toughest mile.
29:38So, again, you're baking this stuff in on top of a great proposal, on top of doing it in literally under five minutes from the moment it was requested. So that's kind of the magic here that we can start to see when you get to this level of AI native operating.
29:55And I think something that's cool is, again, normally, I just fired this in Claude, but it's magic for me when I don't even know a proposal was asked for because I'm on the road.
30:06I'm in meetings. I'm doing something else. An email comes into my inbox, and someone says, k.
30:11We'd love to learn a little bit more. I'd love to see a proposal or show me what this might look like. The brain will understand that, which I'll explain right now, pull in that trigger, and fire this all without me ever having to lift a finger or even know that I needed to do this.
30:26So it's crazy to get that Slack ping and then be like, oh, proposal's ready. What are they talking about? I already have a proposal.
30:32And then I go back and see what their reference of the breadcrumbs were, and I'm like, oh, amazing. I have it. Review.
30:37Send it off. And they're amazed because they're like, how'd you get this to me? I just asked for it.
30:41And, you know, again, you gotta balance that, but I think it's really cool. I'm actually what I'm gonna do is start another one quickly as I talk into context because I don't wanna miss out on this.
30:54And this one, I'm gonna speak to you a little bit, and I hope it works. But this is gonna be I showed you that Spotify prototype. They you saw that proposal.
31:01They wanna increase retention. So I know people use whisper flow flow.
31:07People use a bunch of other things. I'm a Luddite when it comes to this.
31:10I just like using the native mic feature. And to me, I find it works fine. So let's do that.
31:17I wanna create a feature for Spotify. I want it to help increase retention.
31:24I think it should be a daily mini playlist or a daily blitz of three songs. I should be able to access it from the homepage.
31:33And when I get in, there are three handpicked songs for me. I know why they were picked.
31:38I'm able to save that playlist, share it with a friend, or play the music. And the goal of this is to build this in under ten minutes, use all the context you have, make sure it's beautiful and matches the design system, and make sure that I can test retention.
31:58Okay. So I have this. I'm just gonna go backslash goal here.
32:05Because like I said, running this command helps make sure this is going.
32:10I'm gonna run the other skill chain that I was talking about. So we have this skill chain. So I'm gonna run this.
32:17Again, this I'm running it on medium effort and I'm running it on auto. I would never normally do this. I would have it on high effort or ultra high.
32:25I might even not for this episode, but I might even have the workflow feature in now where I have sub agents going and really trying to optimize this design by going and vetting other things, and we're not gonna talk about that now. I don't think. And I would definitely have permissions on.
32:40I would wanna review if it's building the right thing. I would wanna review some of this for this high stakes work. I love autonomy, but there are levels and places where you want autonomy to start.
32:50So while this is building, you wanna talk about context? Yeah.
32:55Let's do it. You're lucky. This is this whole episode should be.
32:58Let's go. No. I just have a big I think this is, like, a such a key part of the whole puzzle.
33:03So, yeah, let's let's go into it. It is. This is that foundational layer that powers the agents to make you truly AI native and therefore your org.
33:11So before Greg, could you tell me what LCA's SOP is for getting back to clients?
33:19No. I cannot. No.
33:21Could you walk into let's take you back, stumble upon and know what their strategy was for 2014 when we were there, 2013, 1415, What their strategy was or their definitions of success were for 2014.
33:37No. I can't. Even though I was in those board meetings.
33:40Exactly. And could you tell me who just got hired at LCA two weeks ago?
33:45I could no. I can't.
33:48No. Yeah. So
33:50even I struggle with some of these things. Right? Because there's so much going on and everyone's doing this and then multiply that when you're at a bigger company by n number of teams and people.
34:01You're essentially blind to the organization. And I think, you know, big reveal, the context layer.
34:08The context layer, like, literally allows you to see everything And not everything in a in a yeah. Exactly.
34:15The eyes open? I didn't check that at all. I know.
34:17I know. We we're storytellers here at LCA, as you know. But the the true magic is, of course, you can add permissions.
34:25You can make sure what's gated. You can make sure that people see the right things at the right times. But what's cool is you're essentially giving agents twenty twenty vision on your company.
34:34And so when you have these questions, when you wanna know these things, when you're building stuff that requires this type of intel, you have it. You don't have to wonder.
34:43You don't have to send 14 messages or wait days for the answer. You have it. And that's the magic of this context letter.
34:49So I'm gonna zoom out and just walk through it at a high level. High level, let me take you through it.
34:55There's a capture stage, a curation stage, storing in your brain or this context layer, using it to execute, and then having customers experience it, and that all flowing back into itself.
35:07So let's talk about capture. You have a bunch of stuff going on in your company from a bunch of different tools, places.
35:14We have Slack messages. We have meeting recordings. We have emails.
35:17We have boards in linear. We have on and on and on it goes. All of this information contains context, and a lot of it is actually really helpful in producing what you need to produce for customers.
35:29So I have a routine that runs to collect this and bring it into almost like an inbox for my brain. So every hour, takes it in, maybe every two, takes it in, and leaves it there.
35:40It brings all this stuff in. And you can give rules. You can say where it pulls from and what folders to look at.
35:45Very easy. And you can just go and build this if you haven't already in Claude. You can jump in and create a routine right here in the routines tab.
35:53And you can set it up. It's a cron job. It essentially just runs regular.
35:55It's like co work scheduled tasks, but on steroids.
36:00So you can run these and work with Claude and create it. It'll do it.
36:04Or work with Codex. It'll do it. So you bring all this stuff in.
36:09You don't want everything in your brain. Right? You don't want all of the information from everywhere sitting in every folder come like, piling up, piling up, piling up.
36:18You wanna curate it to a degree. So before you file it, you have, like, a cure almost . Okay.
36:25Cool. What actually needs to be in here, what do we want in here? So reads it, cleans it up, files it, decides what's to ignore, and then some of those things might be triggers like the proposal I spoke to you about.
36:36What do we act on? So it detects some language, acts on it. So it's a curation step.
36:41And then you store it in this brain or this company readable agent readable context layer, this memory layer, this brain that again, like I mentioned, there's some other companies solving this major enterprise level, other levels like Aglean, for example, Notion AI.
36:59They're like search plus context plus an agent layer chat layer on top. You wanna get locked into that provider, a little bit of a black box on how it all works, but sometimes really great for your use case, awesome. We like to do some of the things ourselves, and this to us has worked really well.
37:14So the brain here is, like I said, it's just a series of folders with a bunch of different files in those folders organized in a way that agents can search and retrieve and then write back to and improve over time.
37:29That's in your brain. You have agents, people managing those agents, pulling from that to execute and do the work.
37:36So they leverage the context. So I think what I covered, you know, you wanna bring in the context, you wanna file the context, you wanna make the context legible, and then you wanna leverage the context. You can direct the agents and set goals for them.
37:48You can ideate and prototype, which is what we're actually doing. Like, right now, it's Cooking and Claude. You can create these artifacts.
37:53You can run skills and tasks. You can review this work, ship it. And then if I zoom out, all of that flows back into the work, into the system itself.
38:03All of those little things, and I'll get into traces in a moment or exhaust as some people like to call it. From that execution, you ship it out.
38:11People get to experience it. The context actually becomes value. And that's what you're trying to unlock as an AI native org is going from that system, using it to work at incredible speeds, and then you're getting signal.
38:23You're delivering value, you're getting signal back from customers. So I'll talk about a labs page in a second on how you can ship some of this stuff out. It can realize the value.
38:32And, again, all of that signal from the market goes back into the system and then gets curated and then back into the brain. One little note I'll add before I, you know, I wanna hear what you have to say or ask about this is a lot of the work that gets done or produced, let's say that proposal or let's say this prototype, there are lot of decisions made along the way, which is tough in big orgs or even in small orgs to keep track of of why did we make that decision.
39:02There's a lot of work that happens along the way, a lot of documents, explorations, etcetera, that that are actually really valuable. So that's like cutting room floor stuff. Those are the traces that's some people call the exhaust.
39:13That's really important to come back in and then make new artifacts based on that. Maybe there's a learning or a lesson in how to get to a decision like this or how to create something. And your brain can act on all of these traces to create this and store it instead of leave it in these this graveyard of files that no one ever looks at again.
39:33So another really cool thing about this system.
39:37So what you don't wanna have happen is you're bringing in the wrong context.
39:43So you basically don't wanna have output that agents are doing and and it flowing back into the capture because, you know, you wanna make sure that the human has basically said, like, this is good.
39:56This is bad. Edit this. Right?
39:58So are is what you're saying, you know, the experience is what is the human layer that basically allows the right type of
40:12context to flow back into the capture section and and and the brain section? It's a great call out. So a couple of things.
40:18One is here, the humans still manage the agents. Right?
40:22So you still have to have some human in the loop and some judgment on what is good and what isn't. And when you do, whenever you're chatting, whenever you're managing that agent, you will be telling it stuff, and it will be remembering it and writing it back, updating the skills, making sure it knows what's good and what's not, updating the memory, and maybe even piling it or packaging up into lessons.
40:42So that's one piece here on the leveraging the context, and this is more from your customers and from the market. So you're gonna see stuff on how they're reacting.
40:51Are they buying more because of the new feature that you just dropped? Are they churning faster because of your new landing page?
40:58All that stuff. And that signal is what'll flow back into your tools, and then therefore, it'll back into your brain, and then update accordingly.
41:06Makes sense.
41:07Let's check-in on Claude. Okay. It's built.
41:10So what I didn't cover briefly is the labs page. I showed you this, but this was part of a labs page.
41:16We should see ah, there it is, the daily blitz. So we spoke about this. This is what we before this didn't exist, it's here now, which is awesome.
41:24But before we had this live event thing that I showed you, let's check out the daily blitz and see how it turned out. Alright. Slow burn.
41:32Sounds funky, man. Let's let's dive in.
41:35So this is actually pretty nice. This is, like, pretty clean card. It's right on the homepage like I asked.
41:42And we can click in. Now you have this playlist. Why we built this for you?
41:48Great. It tells me why. One you love with two fresh picks.
41:50Love that on a little playlist. And I can play this blitz very loud in my headphones.
41:56You probably didn't hear it, but I actually have the music playing live right now, which is super cool. Um, and I can also share it, which is awesome.
42:05Oh, I've got some friends there, and I can share it.
42:09So just for context, the reason why LCA is building these things is, you know, you have two there's two parts of the business.
42:18Right? There's the how to make AI native org stuff, which is what kinda like what you covered, which is like the skills and just helping companies figure this out. But the other part of the business is designing the next iterations of apps, websites that are AI native.
42:35Right? So a big part of your proposal process is I mean, there's a million design firms out there.
42:43Right? So you you wanna stand out. And a way that you're standing out is by sharing these prototypes with stakeholders at at potential clients.
42:55And you're kinda just showing you're using basically, you're doing is you're using all the amazing contacts from the team and all the years of of, you know, six years of work of working with the world's largest companies, and you're kinda, like, putting it in there, and that's helping you kinda inspire what these prototypes look like.
43:15Is that correct?
43:17Totally right. And the new unlock here is how fast you can get feedback from some of the stuff you're producing, which is actually, as you know, in the game of product, the whole game.
43:28Right? Because you wanna produce things and check them out and see how they feel as much as you'd wanna write a however many page PRD and slowly build it and get it out there over weeks or months. If you can build a prototype in under ten minutes that looks like this, allows people to feel it, get real reactions, that's, like, the game.
43:48Well, so my question is is is is is an obvious one, which is it's like, oh, someone listening to this is like, okay. Great, Theo.
43:55You have all this amazing context because you have, like, a team of 55 people or whatever of some of the smartest people in AI and product. I don't have that context. So for people who, you know, don't have that context but who want good, you know, good outputs, be it product, be it whatever, how how do they bootstrap context?
44:20The world is a large and lovely place, my friend. So we are not the only people who have produced beautiful work.
44:28I think for net new stuff, we're among the best in the world of thinking about AI flows, conversational UX, how to design for trust, especially with agents. Not a lot of people have done that.
44:39If you're looking back on this, this is go to Mobin. Mobin has an MCP now. Mobin is a library of a bunch of beautiful apps, their flows, all of the different permutations.
44:50Get Spotify's design system or another one that's similar. Create a skill around it. Plug into a Mob and MCP, and all of a sudden, you can create this in minutes as well.
44:59So this isn't only LCA stuff. Maybe the idea okay. Cool.
45:02Retention. You know, we had something called the daily five back in an early startup. And maybe these ideas are easier to come by for us or faster to come by from us, and maybe our agents are more plugged into that.
45:14But in terms of producing something like this, people can do it just by using the right tools and creating the right skills and then slowly loading in the right context over time. Right. So I think the the takeaway there is once you figure out what your output is,
45:29you wanna see, like, which MCP exists for that output and then see how you can kinda scrape some of these ideas and and things that are, you know, working or trending and stuff like that such that the, you know,
45:49the output is good. Exactly. So those are the tools piece.
45:53Yeah. And you give them the skills piece as well. So create a skill around the Spotify brand or whatever company brand or new brand.
46:00Maybe there's some great there are a ton of great UI skills out there as well. You give it a clear goal, and then the context you have, well, maybe you're light on that if you're going from scratch, but find ways to provide context on why this would be a great product or what would make it a great idea, and then feed it those MD files or kinda give it that access.
46:22This is something that I wanted to do with you. I know we're we're coming up on time, but maybe if we can, let's let's do it. So on the labs page, what we didn't show is this test.
46:32Right? This is just the the labs experiment is part one. The test is part two.
46:37You can flash your phone now, Greg, and do this if you want, or I can just copy this URL and send it to you. But I'm gonna show you what this test looks like, and I'm gonna Slack you the URL if you're cool with that.
46:49And what we're gonna do, I'm gonna complete this test. So right now, this is part of the skill chain that I was talking about earlier.
46:57So we have a skill chain firing for this that essentially looks like these five skills right right away. There's a hypothesis we're trying to test.
47:05You know, we cruise, we blitz through it in Claude code. Normally, like I said, I would go through these. Then the build prototype skill, then a usability test skill, which we're gonna show right now, a feedback synthesis skill, and then a v two skill.
47:17So this was cool. Imagine getting feedback right away and building it on the spot. That's even cooler.
47:23And so that's what these skills allow us to do. So if you have it open on your, excuse me, on your screen right now, you can start it.
47:33And this is essentially what you're gonna go to. There's a little bit of a usability test. This is, like, what a researcher might do with someone.
47:40And it asks you a few questions. How often you listen to Spotify? How do you find your music?
47:46I replay what I know. And now I can open this daily bliss and it tells me, like, high level what to do. I go through the workflow.
47:54It asks me questions along the way. You know, how much did you wanna listen to these after looking at the songs?
48:01A lot more. I love them. Go through it, etcetera, etcetera.
48:06And then I'm just gonna kinda quickly how would you like, uh, very likely except I wish it had more more options or something like that.
48:19Mhmm. How valuable? Let's just give it a five.
48:23Actually, let's give it a seven. I need new music always.
48:31How easy is it to open? Let's just keep going. Yes.
48:35I did. Great recommendations.
48:41Can you say more? No. I can't.
48:44Okay. So not sure if you were able to fill it out also, but that was the prototype.
48:51I just filled in essentially a a research report. You can see the signal tab right now.
48:57There's zero out of 10. Oh, one completed.
49:00That was me. I just completed it. In theory, you could send this link to five people, 15 people, 40 people, whatever you want.
49:09Maybe you have a community on Discord or Slack that you want early testers. You send this out. People complete it, and then all you have to do is now with one answer, we're probably not gonna get a great synthesis.
49:20But all you have to do is click this. This is another skill, and it'll synthesize the results. Imagine when there's 50 results or a 100.
49:27Oh, actually, you do. Here's some lessons. It generates some lessons.
49:30One expectation, discovery, validation. And right there, I could click plan v two and then execute it,
49:37and I could have a v two done in the same session. Wow. And then I think, you know, I've seen some people talking about, like, autonomous product building and stuff like that.
49:45Like, that's where this comes in. Right? So on this on the Startup Ideas podcast, I talk a lot about building companies via the ACP framework, audience, community, product.
49:56And in an AI world where this exists, you know, your product, it just like you you have this deep connection between the community and the product.
50:07And you're just able to create a a product that has a high higher probability of success when you have something like this. Right?
50:16Otherwise, you're just kind of flying blind.
50:20Totally. And this, like, the fidelity of the insights that you get, again, that perfect twenty twenty vision because of the context is awesome. So you can go to results, and, eventually, you can see everyone here.
50:32This is a little custom thing that we built, but, again, did it all with Claude. Wasn't that challenge? It's not like, you know, it was all elite engineers doing this.
50:39We were able to do it, and then you can have a report generated. And when 50 people, 20 people, 10 people have kinda gone in and tested, you've got that.
50:48So I think a really cool opportunity. You can see the speed impact on speed. There's a little grid here.
50:53We don't have to talk about them all. But a proposal might have taken up to three days, and now it takes minutes. You can saw that clickable prototype.
51:00Again, not a prototype in Figma, not a design prototype. A functional prototype could take one to two weeks to figure out what to build, do it, get it out in people's hands, took minutes, not only to get the prototype done, but also collect feedback and then potentially synthesize into the second version.
51:16So a lot of cool stuff. If we have five minutes, Greg, do wanna jam startup ideas, or are we out of time?
51:24Let's give a let's give a few startup ideas. And and, yeah, I'm curious what what you got.
51:31Okay. I want your feedback here. My take is based on what we just showed you and based on what we just talked about, this AI native system of people, agents, and context that unlocks speed for companies that gets them signal in real time, allows them to build better things, and create a moat.
51:50This is a framework that we love, that we created, that we use. You can now go deploy this if you want in what I think is the hottest, best market right now for startup ideas if you're into services.
52:03And, eventually, you could create products. So TBD, if it's a thirty day sprint or an AI acceleration team, you're very incredible with offers.
52:12But the game here is to niche down, which you always talk about. And the three vectors is in this our industry function and company size. So industry could be pick your niche, commercial real estate, dentistry, whatever it is.
52:26Pick it. Restaurants are very hot, very, very hot niche right now because they are especially fragmented and can really use this.
52:34Now you can't go too small. They won't have the budget. But as you go up, function, who do you wanna support in that in that industry, which team, and then company size.
52:43And then get incredibly good at understanding those workflows. You might already have an unfair advantage in one of these and producing the right service offering to help bring the system to those companies.
52:56Does that track?
52:57I mean, yeah, this is, like, no brainer. This is, like it's it's it's stupid how good it is.
53:04You know what I mean? It's like
53:06yeah. I if you're like, hey.
53:09What can Latecheka do? We're gonna spin up five new companies. It would literally be this times five with different industries.
53:14Like, that's literally what do. Yeah. I think, like, LCA in theory,
53:18like, would do this. But because, like, LCA focuses on Fortune 2,000, there's just so many other markets that people can go after.
53:30Yep. And go for it. A way to prioritize it, you had a similar to upgrade in your newsletter, which I I loved.
53:39I changed it just slightly to go from niche to general and then low frequency, high frequency. So if you can find niche workflows, so force very specific niche, industry, function, company size, that are high frequency, and you can create those workflows and show people those on a sales call, in a brief, in a proposal, in content.
54:02Keep doing that over and over. You will, you know, you will have a layup ahead of you. And then you can go to general, but still very important because once they get the niche stuff, they're gonna want the stuff that they do all the time that's a little less niche.
54:14And then you can go into the high value niche but low frequency but might have higher ROI.
54:19Love it. Cool, man. And that's the episode.
54:23Right?
54:25We could cook for days, my friend, but that's the episode for now. Yeah. I think there's we can go so much deeper into so many
54:33parts of this. But in under an hour, this masterclass of how to become AI native, showing some examples, this has been amazing.
54:44I'm putting you on the spot, but I'm gonna include as my pin comment, if you're a company doing more than $10,000,000 a year in revenue and you're looking for a free consultation from Theo or team, Nice.
55:01Yeah. You can go and and grab it. Maybe you can do give away like 10 or 15.
55:10Absolutely.
55:11Maybe not 15, but we'll give away a few for sure. Okay. We'll give away ten ten, fifteen minute consultations.
55:17If you're a company doing $10,000,000 a year of revenue, go and click the link. Theo is a criminally underfollowed, uh, account on on on the on social and x.
55:29I'll include where to find him on on the Internet too. Theo, is there one thing you wanna leave people with for this episode?
55:40I think, like, all this stuff about AI native and AI everything can be overwhelming and can sound like a lot, and it feels like you have to be a technical guru and genius to just get started and just kinda make dense in this progress. But, really, to become an AI native org, think through the lens of managing agents and what those agents need to succeed,
56:01and you will be well on your way to being ahead of most companies in the world. So I think just get started. Don't be scared.
56:08Scrape your knee and get stuff done. And and if this has been interesting, just let me know, you know, because, uh, I'd love to have Theo back on the podcast again, but I wanna create stuff that is valuable for you. So please let me know in the comment section.
56:22Like this con like this episode if if you got an ounce of value out of it. And I'll see you in there. I read every single comment.
56:29I respond to a lot of them. And, Theo, I hope well, I'll see you I'll see you soon, but I hope people like this episode and, uh, you come back on again.
56:39Thank you so much. Too. We'd love it, man.
56:40Thank you. Cheers.
The Hook
The bait, then the rug-pull.
The question opens in two seconds flat, no preamble. By the time the first guest appears, the episode has already promised something the internet usually charges tens of thousands of dollars for, and then it actually delivers.
Named ideas worth stealing.
06:57model
The AI-Native Org Definition
- People manage agents
- Agents read and write to the company
- The company gets smarter over time
The three-bullet definition that separates genuine AI-native orgs from companies that merely use AI tools.
Steal forAny positioning deck, sales call, or consulting intake to benchmark where a client currently sits.
15:09list
Four Requirements for Agent Autonomy
- Goal (specific, measurable, timely)
- Skills (playbooks, SOPs, reference docs)
- Tools (MCP, internal, external)
- Context (the company brain)
What an agent needs to run without constant hand-holding, analogous to what any new employee needs on day one.
Steal forOnboarding checklist for any new agent or workflow; diagnostic for why an existing agent keeps failing.
34:54model
The Recursive Context Layer
- Capture: hourly cron from Slack, email, meetings, Linear
- Curate: librarian agent reads, cleans, files, ignores, or triggers
- Store: organized markdown brain, agent-readable
- Execute: agents leverage context to produce work
- Experience: customers get value, signal flows back in
The five-stage loop that makes an organization machine-readable and gets smarter with every interaction.
Steal forBlueprint for building any company brain or AI-native knowledge system from scratch.
52:08model
Verticalized AI Acceleration Services
- Industry (real estate, dental, restaurants, logistics)
- Function (sales ops, support, claims, recruiting, finance)
- Company Size (under 1,000 units / under 50 people / single location)
Three niche vectors for selecting where to deploy AI acceleration services.
Steal forPositioning a new consulting or productized service in the AI-native space.
53:40model
2-Up Prioritization Map
- Niche + high-frequency: start here, show on sales call
- General + high-frequency: expand here after niche is won
- Niche + low-frequency: high ROI, build after proof of concept
- General + low-frequency: last priority
A two-axis map for sequencing which AI workflows to build and sell first.
Steal forRoadmapping which use cases to tackle first in an AI transformation or productized service.
How they asked for the click.
VERBAL ASK
54:42product
“If you are a company doing more than $10 million a year in revenue and you are looking for a free consultation from Theo or team, go click the link.”
Greg puts Theo on the spot live; Theo agrees to 10-15 consultations. Pinned comment CTA. Low friction, high qualification bar ($10M ARR floor).
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
