Latest / Lenny’s AI Builders / From 3 to 12 AI Employees - The Context Engineering System
Transcript
- 0:00Speaker 1: Guys this is the future. This is what is now possible with AI Agent, Open Claw and self improving loops. I can't code. I have 32 ks in savings and last week I gave 12 AI agents control of my entire business. One of them opened eight code changes in my app overnight without me asking.
- 0:18My co founder reviewed them and approved all eight. In the next twelve minutes I'm going to show you the exact system, the second brain, the 12 employees and what happens when you let AI agents run everything. About ten days ago I made a video about OpenClaw, it's the open source AI agent framework. Peter Steinberger, he's the guy who built it, he just joined OpenAI and that video did about 3,000 views on my channel which at the time had about 700 subscribers which is insane. Peter said something in an interview recently that rewired how I think about all of this.
- 0:48He said Agents will replace 80% of apps, not because they're smarter but because they have your context. Your fitness app doesn't know you slept badly. Your calendar doesn't know you're stressed but an AI agent that lives on your computer knows everything. And that is the difference between a chatbot on one hand and an employee on the other. An employee knows your business, your audience and your numbers.
- 1:10And that difference teaching your AI about you instead of just talking to it that's what people call Context Engineering. And I know Context Engineering sounds like another tech buzzword but hear me out. It's actually the simplest concept in the world. It's just what information does your AI have when it makes a decision. So if you give your AI a prompt and nothing else you get just shit.
- 1:34It's generic crap. But then if you give your AI a prompt and your voice memos, your reading highlights, your daily notes, your customer data, your performance metrics you get something that sounds like you, that thinks like you and eventually acts like you. There is a creator called Vin who said something on Greg Eisenberg's podcast. He said Stop managing the agent, manage the vault. Your agents just read from it.
- 1:59So if they make bad decisions you're not fixing the agent, you fix the vault with all of the information in it. That actually really clicked for me because that's exactly what I built without realising it. And here is the thing, if you're creating content right now you already have context, you have ideas, you have experiences, you have things you want to say. The problem is getting all of that out of your head and then into something useful. That is exactly why I built Threatify, takes your raw ideas and turns them into content that actually sounds like you, not AI.
- 2:28Threatify. App is the link, it's in the description, you can try it 100% free without paying a cent. No credit card required. Check it out. Alright let me show you what happens when we take this idea of context engineering way further than just content.
- 2:41It's pretty fucking cool. This is my second brain. And before you think it's some complicated Notion setup with 58 different Notion databases it's not. It's literally just folders with text files. Super simple.
- 2:52Each folder is a year and then each year has the 12 different months in it. Each daily markdown file contains pretty much everything I've done in that day because I'm logging all of my work. Here is where it gets crazy though because I am constantly recording stuff. I'm recording conversations with my girlfriend, every piece of content that I consume whether it be a YouTube video, a podcast, a book, an article anything, all of the highlights from that are taken and put into this Second Brain system because I use Readwise. All the daily notes I take, I just put it into my OpenClaw and ask the OpenClaw to log it to the Second Brain file.
- 3:28Then we have the performance metrics, the customer data, all the stuff that we need to actually have this vault of information like I was talking about a moment ago. It just flows into the system. Now here is where it gets wild because I'm not the one reading these files anymore. 12 AI agents review them. Every single morning before I wake up, my agents read the second brain.
- 3:48They know what I highlighted yesterday. They know what content I consumed. They know what I was talking about with my girlfriend because those voice memos were recorded. They know what the MRR of the business is, Threadify, they know how many subscribers I have, what my last video's retention curve was like, what the open rate for my newsletter was like. All of it.
- 4:06Let me walk you through them. I have 12 AI employees each one has a name, a personality, a constitution which is like their operating manual and performance reviews written by another AI agent. Firstly we have Remy Remy is my researcher. Every time I want to make a YouTube video like this one Remy scans the last thirty days of trending topics from Reddit and X, checks what competitors are covering, reads my Readwise highlights, minds my second brain for personal stories and then produces a Topic Brief. It's not just here's a list of ideas it's a full brief with talking points, competitor analysis, readwise quotes scored by relevant and personal anecdotes from my own voice memo.
- 4:45Then we've got Max. Max takes that brief and brainstorms 10 title options for a YouTube video, scores every single one against the constitution of title writing principles, then picks the top three as Title thumbnail pairs for AB testing in YouTube. Each title comes with a thumbnail direction for the thumbnail agent. Then in the pipeline we go from Max to Harry. What Harry does, he writes five hooks using different psychological principles about hook writing, pattern interrupt, curiosity gap, bold claim, vulnerability, data, things like that, scores them all, then picks the winner without any human review.
- 5:19The hook you heard at the start of this video was written by Harry. Then Sage and Teresa, they run-in parallel at the same time. Sage writes the full script. This script, the words coming out of my mouth right now using Harry's hook word for word as the opening, and then Teresa generates the thumbnail options that are matched to three title pairs. Then we have Enrique, one of my favorite employees.
- 5:41Enrique designs the diagrams and visual models that you're seeing on the screen. Everything inside of ExcaliDraw, these six employees, Remy, Max, Harry, Sage, Theresa, and Enrique, that's all in the preproduction phase. And then the human in the loop, me, comes and records a video. Alina, who is over here, Alina will timestamp everything. Nina then writes the newsletter that's repurposed from the YouTube script.
- 6:07Piper the publisher then handles publishing. Then Simon who's in purple over here takes everything that just happened and synthesises it back into the Daily Second Brain file so that the next time Remy the researcher runs Remy has even more context. Lastly there is then Aria. Aria is the analytics engine. She reads every metric views, retention curves, subscriber conversions, newsletter open rates and then writes performance directives that every other employee reads before they start work.
- 6:38These performance directives could be anything from Vulnerability hooks are outperforming bold claims (two:one) or Data stacking titles with two or more numbers drive 3x more clicks. Basically every employee gets smarter because Aria is watching those numbers. Now all of this is cool but the moment that broke my brain was Gary. Gary Growth is my autonomous growth agent. He's not part of the content system or the pipeline that I was showing you a moment ago.
- 7:03Instead Gary looks at my entire business and finds ways to improve it. A few days ago I woke up and Gary had opened eight pull requests in Threatify's codebase. Eight blocks of actual code changes, tracking fixes, safer routing, rollout controls, pricing clarity, experiment configurations. Things I didn't explicitly ask for, honestly things I wouldn't have known to ask for. So my co founder, my technical co founder Sam Sam he reviewed every single line, approved all eight of them.
- 7:31The code was good. An AI agent who reads my second brain wrote real production code in my app while I slept. I literally cannot code and my AI agent is shipping features. So where does all of this stuff actually land? Well Threadify is $17.14 dollars in monthly recurring revenue, my YouTube channel has seven forty three subscribers and it's $0 in custom code powering this entire pipeline.
- 7:5732 ks in savings, ten months of runway, this is real, this is everything I've got. Remember what I was saying about that dude Vin? He said Writing is how you delegate to agents. If you develop a writing habit you have way more context to pass to an agent which dramatically increases what you can delegate. And that's exactly what happened with me.
- 8:15My voice memos, my content highlights, my daily logs. It's not just journaling, this is all fuel for my 12 employees. And here is where all of this goes next. My girlfriend said something that I keep thinking about. Once you have Gary growthed every single system in the business then you can sell something like Gary for other people's businesses.
- 8:36An AI agent that is dedicated to one thing growing your business. If that is something that you can actually reproduce and then share with other businesses that agent is priceless. Okay. Look. I know all of this sounds like a lot.
- 8:5012 agents, 82 skills, second brains, but I I didn't start here. Six weeks ago, I had zero agents. So if you want to start, here's what I recommend you do. Firstly, start capturing everything. Voice memos.
- 9:02Set up Readwise so that it connects with different sources of information that you consume. It doesn't matter how messy all of this is you just need to get it out of your head. Step two: Pick one agent and then just give it one job whether that is research like I have Remy or Titles which is Max or Hooks which is for Harry. Just pick one use case that you are sick of having to do manually and then create an AI agent to do that for you. Three: Then feed it your context.
- 9:30Not generic prompts like a chatbot but those voice memos, the customer feedback, the metrics. We are not having a chatbot which is generic. We are having an employee. And the difference between a chatbot and an employee is context. Step four: Let it run for a week.
- 9:47Review the outputs it is giving you. Fix the vault, not the agent. The agent is only as good as the information you give it and that information is context. Give it everything. Step five: Once that first agent is running you're pretty happy with it add a second one.
- 10:02Connect them. Watch what happens. This entire pipeline feeds into one product Threadify. Every improvement Gary makes, every insight that Aria surfaces, every newsletter that Nina writes it all exists to make Threadify better for you. The same AI that writes my scripts also analyzes what my audience wants.
- 10:20And that's how Threadify gets better every single week without me writing a single line of code. Quote Glenn, one of our power users, he said After four posts barely any editing needed. And that is because the knowledge base inside of Threatify is overpowered as fuck. It's got all of these facts about you so that you can literally just prompt your own content employee inside Threadify and it will automatically incorporate the way you write because it knows facts about you, about the work you do, about your audience, about your writing style and about your offers. That is what context engineering does when you give your AI your real context.
- 10:53Try Threatify for free. There's no card required. Link is in the description. Just go try it. I post three or four videos a week, Tuesday and Saturday, bit more of polished vibe.
- 11:01And then on Wednesday, there is a livestream where I'm building Threadify live. Subscribe if you wanna watch a non technical founder with thirty two k and ten months of runway. Try to get to 10 k a month using AI agents like the ones that I've shown you today. If you haven't seen part one, the Open Claw video that started all of this, it's linked in the description. 3,000 views.
- 11:22Got a 100 subscribers from it. It's the free quote through everything that you've just watched. Don't be a dropkick. Subscribe.