Two Pioneers. One Pizza. 1 Owner. The new lifecycle of a feature in the agentic era.
A single new feature used to require twelve people. A PM, a PMM, a PD, three engineers, a data scientist, a researcher, a tech lead, a designer's manager, the PM's manager, plus the meetings that come with them. Six standing meetings a week. A status doc nobody reads. Six months minimum.
The same feature today takes two human crews and a pack of agents. 2 Pioneers validate the wedge. 1 Pizza Eng scales what the pioneers proved. 1 Owner + Agents maintain and enhance it for the rest of its life, the agents run ops, the Owner ships 80% of enhancements alone. AI absorbs everything between.
This is what an AI-native product organization actually looks like at the unit level. Not a roadmap. Not a headcount plan. A lifecycle.
The whole lifecycle, on one page.
Three crews. Three jobs. Same feature, moving from validated to scaled to maintained. The headcount actually shrinks as the feature matures — that is the AI-native shape.
Two domain owners. PM, PD, Eng, Data, business owner.
Whoever’s closest to the problem. Owns the wedge, the customer evidence, the kill criteria. Runs interviews. Drafts the thesis. Ships a 30-day MVP with pre-committed design partners.
Output: an alpha 5–20 partners actually use.
Tools: Claude Code, Cursor, prep doc per call, thesis with explicit kill criteria.
3–4 engineers with AI in the loop.
Takes the alpha and hardens it for production. Integration, infra, observability, security review, rollout plan.
Output: a production-ready feature with metrics instrumentation wired up.
Tools: Agentic coding, AGENTS.md committed, CI with eval gates, on-call from day one.
1 person of any role + Agents.
Often a Pioneer staying on. Agents watch usage, catch regressions, draft the health report, file routine PRs.
The Owner ships 80% of enhancements directly with Cursor or Claude Code. The hard 20% routes back to engineering.
Tools: Observability with eval gates, agents with scoped write access, published runbook, Owner with repo access.
Note the shape: validation needs taste. Scale needs craft. Maintenance needs coverage — and coverage is what AI is best at. The old org chart kept all three crews staffed for the life of the feature. The new one collapses the back half.
Six waypoints. Six jobs. None of them are "review the deck."
"AI absorbs everything between" is the load-bearing claim. Here is what that actually means.
That is the twelve-people org chart. Each of those roles existed because each of those workstreams was a human-shaped job. They are not human-shaped jobs anymore.
The work that survives is the work AI can't do yet. It clusters in three places, one per crew.
Validation, scale, maintenance-and-enhancement. Three crews, one line. The headcount goes from 2 → 4 → 1 (plus a pack of agents doing the rest).
Most CPOs ship features. The Builder Factory CPO ships the line that ships features.
The main goal is your factory’s time from idea to revenue. That single number is what the whole line is judged by, and the only way to drive it down is to instrument everything along the way: every gate, every handoff, every crew. If you can’t see it, you can’t shorten it.
Two Pioneers validate. One Pizza scales. A pack of Agents plus one Owner maintain and enhance. AI absorbs everything between. The CPO’s job is to enforce the gates, instrument the line, and protect the crews at every waypoint.
The old org chart was a list of twelve people. The new one is two pioneers, four engineers, then an Owner and a pack of agents.