Latest / Lenny’s AI Builders / 12 AI Employees, 1 MacBook, $0 Customers (Context Engineering)
Transcript
- 1,043 views in 2 days. 12 ai employees. one macbook. a self-improving system that rewrites its own rules after every video. and zero - i mean zero - new customers from any of it. let me show you the system, the numbers, and why the smartest ai pipeline i could build still can't convert a single viewer into a paying customer. so this is my youtube studio right now. that video - "12 employees. 8 prs overnight. $0 in code" - hit 1,043 views in 62 hours. that's my second best launch ever. +15 subscribers in a single day. best daily growth i've ever had. and this is my threadify dashboard. same time period. zero new email signups. zero free trials. zero paid conversions. nothing. $1,753 a month in recurring revenue. sounds like progress until you realise my take-home after all costs is $293 a month. and my gross revenue retention just dropped to 62.7% - meaning i'm losing over a third of my revenue every single month. but here's the thing that keeps me up at night. this video you're watching right now? it was produced by 12 ai employees running on this macbook. the system that made the breakout video... is the same system making this one. and it still can't figure out the money part. let me show you how all of it works. everyone's still calling it prompt engineering. writing better prompts, refining your instructions, getting the output right. that era is over. martin fowler - the software architecture legend - just published a framework called context engineering for coding agents. google released ADK with tiered context architecture. the whole industry shifted in the last month. context engineering isn't about writing better prompts. it's about building better environments for ai to work in. context is everything surrounding the ai at decision time - its memory files, past conversations, performance data, constitutions, rules, examples. this is what mine looks like. 13 separate skill folders. each one is an employee with their own brain. their own memory. their own rules about what to do and what not to do. think of it this way. prompt engineering is writing a good email to an assistant. context engineering is designing the entire office - the filing system, the procedures manual, the performance reviews, the shared knowledge base - so your assistant can do their job without you explaining everything every time. alex finn runs his agent factory on two mac studios. one running anthropic, one running a local model. separate machines, separate agents, separate contexts. i run mine on one macbook pro. the difference? my agents aren't inference endpoints. they're specialised prompt pipelines. and that's why this works for a non-technical founder. quick note - if you're building with ai and your content funnel is broken too - this is exactly what threadify automates for your threads content. context engineering for your social media. 7-day free creator trial in the description. someone in my comments last week - kevin jeppesen - said something that sparked a debate. he said running multiple agents is bloat. one agent can do everything. i completely disagree. and here's why. this is the actual pipeline. employee zero is bevan - the intelligence synthesiser. before anyone writes a title, a script, or a thumbnail, bevan has already pulled analytics, mined youtube comments, read my business metrics, and asked me questions only a human can answer. then remy researches the topic. max generates metadata - 8 to 10 title options, scores them, presents the top 3. harry writes 5 hooks using different psychological strategies and auto-selects the winner. theresa designs 3 thumbnails paired to each title. sage writes the word-for-word script. enrique creates visual diagrams. all on this macbook. all running through claude code. no api setup. no docker containers. no coding. naval ravikant says collect geniuses, don't hire for roles. same principle here. collect specialised agents, not one mega-agent. each employee does ONE thing exceptionally well because their entire context - their memory, their constitution, their performance data - is built for that one thing. a general agent trying to write hooks AND scripts AND metadata is like hiring one person to be your researcher, copywriter, and designer. sure they can do all three. but they won't be great at any of them. now here's the part that genuinely blows my mind every time i run it. jason calacanis was on all-in this week talking about his team training agents. he described an agent that goes out every saturday, reads articles about thumbnails, finds a mr beast employee talking about heat maps, and adds it to its own skill. every week it gets better at thumbnails without anyone touching it. i have that. but mine is way more aggressive. meet aria. employee number 11. after every single video, aria pulls the actual performance data from youtube, kit, and transistor. views, retention curves, subscriber conversion, newsletter open rates. then she writes directives for every other employee. right now harry's performance file says "DO data-stacking hooks - they retain 35 to 39 percent at the 10 percent mark." that directive didn't come from me. it came from aria analysing 4 retention curves and finding the pattern. the system is teaching itself. alex finn's agents improve the code. my agents improve each other. aria rewrites the directives that harry reads before writing hooks. the system gets better at getting better. and my agent gary growth? he's doing the same thing for threadify itself. yesterday he opened two pull requests autonomously working on connect rate optimisation. an ai agent reading my second brain and writing real production code. without me touching anything. and now here's the honest part. the part that keeps this from being just another "look what ai can do" video. this is my conversion funnel. real numbers. 10,000 plus views per week. zero email signups. zero free trials. zero new paid customers. the pipe is completely broken between attention and revenue. $1,753 mrr. 40 people paying $49 a month. gross revenue retention at 62.7 percent. that means i'm losing 37 percent of my revenue every single month to churn. 13 people have cancelled all-time. my ai employees can produce a polished video pipeline in 45 minutes. they can research, write, design, and publish faster than any solo founder on the planet. they can't make a single person click a link in my description. $293 a month take-home. $32,000 in savings. 10 months of runway. and the best content system i've ever built... has a zero percent conversion rate. but here's what makes this different from a failure story. the system is self-correcting. aria flagged the conversion gap. nina - my newsletter employee - is already testing new cta patterns based on aria's directives. the pipeline IS the solution. it just needs more iterations. jason calacanis calls them agent maestros. david sacks says it's the biggest career opportunity in change management. i'm a non-technical founder doing exactly that - and i'm showing you how, live, with real numbers. the answer isn't more content. it's better context. that's what context engineering actually means. not better prompts. better feedback loops. and mine is running right now. if you're building something with ai and your conversion funnel is as broken as mine - or even if you just want to see how a non-technical founder runs 12 ai employees on one laptop - try threadify free for 7 days. link in the description. "nearly doubled my followers in 10 days. 210k views." - traci tournoux. that's what context engineering does for social media. subscribe. 3 videos a week - tuesdays, thursdays, saturdays. real numbers. real building. zero fluff. and don't be a dropkick. ship something today.