Last week I asked a team of AI agents to act as my product team.
It started as competitive analysis. I had agents tear down three competitors, map my feature gaps, and figure out where my differentiation really is.
Analysis is usually where this stops. This time I wanted to see how far down the pipeline the work could go before a human needed to touch it.
So I built a small product org out of agents. A product manager to write the feature specs. A UI designer to produce the mockups. An architect to poke holes early: vague requirements, missing specifics, feasibility.
That's how I've watched strong human teams work. The architect isn't there to plan the build yet. He's there to find the holes before they get expensive.
The product manager took the analysis and picked 20 features in four buckets. Five to build a moat. Five to convert a competitor's customer. Five for retention. Five for table stakes, the gaps everyone expects filled.
I didn't want to just chase the competition. If you're copying, you're always behind. The buckets are what kept the work strategic instead of imitative.
Then one /loop command set the team looping through all 20: spec, mockups, architect review, revision, iterate. It ran on its own until it maxed out my daily credits on Claude's 200 dollar max plan.
The first pass came back with a miss, and it was my miss. I never told the team the features had to extend the existing product. The specs came back loosely coupled to the app, with a UI that didn't match.
So I ran a second loop: /goal retrofit every feature to the current codebase and current UI. It finished without complaint. A human team would have called that churn, and I would have heard about it for weeks.
In hindsight, the unconstrained first pass was a gift. The product manager imagined features without being anchored to what exists today, which is exactly what most product teams struggle to do. The retrofit made them real.
For the deliverable, I refused 20 dense documents. I had the product manager present to me like I was the executive: a 24-slide deck, each feature with the why, the advantage, and the mockups behind it.
I continue to be awed by the work these agents deliver. Twenty features specced, mocked up, and argued for, at a quality I would proudly put in front of any executive. Done overnight.
The humans still own the ends: I defined the problem, and I decide what gets built. But the middle ran without me. And the strategy work surfaced features I wasn't even thinking about, like migration tooling to move a competitor's customer over.
The real change is this: the more clearly I understood my moat, the less I feared the competition. Now I'm spending my time building it.
This is what I build for clients too. Your own product, built in weeks with a team of AI agents, or your engineering team moved to AI-native development so they ship this way themselves. If you want to see what that looks like in your business, reply to this email or grab time on my calendar.