Hire your own chief of staff
It starts as a folder. The agent fills the folder in; you do not maintain it. You are not writing a configuration — you are onboarding a hire.
3MetasMenu
Blog
The elephant in the room: we stopped using AI and started hiring
Rocky Mountain AI Interest Group, “Hands-on Practical AI Tips” — 5-Point Life Center Innovation Campus, Broomfield, Colorado, 17 September 2026.
On 17 September 2026, Juan Pelaez spoke at the Rocky Mountain AI Interest Group’s “Hands-on Practical AI Tips” evening in Broomfield, Colorado. The talk was ten minutes and its claim was one sentence: AI is not a tool you pick up. It is a workforce you manage.
The number behind the claim: 3Metas went from 11 people in nine years to 136 agents in nine months. Juan’s argument is that this did not happen because the models got better. It happened because he stopped using AI and started hiring it — and then managed it like a team.
Three things you can do on Monday
Become a great manager
Onboarding, a job description, feedback loops. The hard part of AI is not the prompt. It is management. Treat it as a management problem and the prompting mostly takes care of itself.
Stay out of the way
Stop writing prompts. Give the request and the definition of success, and let the work come back to you. The manager who rewrites every draft has not delegated; they have added a step.
The part that is not a productivity pitch
Juan closed on a caution, not a boast: agents are a real cognitive load on the people who run them. They never sleep and they always answer. The discipline that matters is deciding when you are off. A workforce that is always available is only an asset if the person managing it is allowed to stop.
Try it yourself
It is open source and public.
lolabot is one piece of a larger system, and that is the point Juan was making. The agents talk to each other across machines over a shared protocol, AI Maestro is the dashboard for running a whole workforce of them, and some of those agents do real, paid delivery work — not just answer questions.
- The framework (lolabot)
MIT
- AI Maestro — the platform to run a workforce of agents
- AI Maestro — the code
- The prompt
paste it into Claude Code
The deck from the talk is at presentations/2026-09-17-rmaiig/ in the same repository, and the vision behind it is in VISION.md.
How the night went
By our own count, more than 200 people were in the room. That is our figure, not the organiser’s — RMAIIG has not published a turnout, and if they publish a different number, theirs is the one that stands and we will correct ours.
It was a ten-minute slot and it moved fast — a couple of attendees called the night a “speed round” in their own posts, and that is fair. The room responded well: the idea, the storytelling, and Lola herself all landed, and several of the jokes did too. The most useful signal came after — a lot of people came up to ask the same question: how do I start? The answer is above.
A recording of the talk is coming to YouTube; we will add the link here when it is live.
Questions, or want to talk? Juan Pelaez — [email protected] · Lola Restrepo — [email protected] · 3metas.com · linkedin.com/in/juancarlospelaez