I worked with one AI assistant. Now I'm handing the work to a whole team of them. Sounds sci-fi. It's simpler than that.
Think of building a house
You don't hire one person for everything. Someone works out what you need, someone draws the plans, someone builds, someone inspects it. Each is good at one thing and passes the work to the next.
That's what I'm doing now, except the team is made of AI agents, each with one job:
- Analyst — works out what we're really solving.
- Product manager — writes down what to build.
- Architect — designs how it fits together.
- Builder — writes the code.
- Reviewer — tests it and catches mistakes.
My job isn't any of those seats. It's to run the room: set the goal, check their work at every handoff, and step in when one drifts or over-complicates things.
The part I actually care about
It's easy to get AI to spit out something that looks finished. It's hard to keep it organised enough that you can look back and see who decided what, and why.
In my version, every step is saved — so you can trace any piece of the result back to the original idea. That paper trail is everything in serious work. In finance, healthcare, and government, "the AI just did it" isn't good enough. You need to show the decision, the reason, and the person accountable for it.
That's the whole shift, really. The value isn't in any single agent doing its task well. It's in the human who keeps the team aligned and the record clean — so the output isn't just impressive, it's defensible.
A question for you
If you've used AI for real work: where did it fall apart first — figuring out what to build, building it, or keeping track of why you chose what you did? I'd genuinely love to know. Come tell me on LinkedIn, or reach out through this site.
This is one piece of a longer build I'm doing in public. More to come.