When AI agents multiply delivery capacity, the constraint does not disappear. It moves upstream: deciding and specifying what to build becomes the pace-setting limit.
In the roadmap I planned for PMRadar, execution was designed in releases of roughly two weeks, with AI agents working in parallel across workstreams. That inverts the classic assumption about engineering capacity, and the conclusion was direct: this half's bottleneck is discovery, not code. A single PM, with agents running in parallel, can sustain that pace under one explicit condition: discovery (flow maps, specification) has to run ahead of execution, not behind it.
That changes how you govern the process. Every new feature is only born once the flow and the spec are ready, before any code: the decision artifact became the scarce input that feeds the delivery machine, not the other way around.
The generalization follows classic bottleneck logic: when delivery capacity multiplies, the constraint does not vanish, it moves further back. The nature of product work changes: instead of competing for space in the engineering queue, the PM starts maintaining a stock of ready decisions (flows, acceptance criteria, hypotheses already validated) always ahead of the agents. Planning capacity turns into planning discovery throughput.
One signal in that direction: in a recent mentoring session, a group of PMs in career transition reached the same conclusion independently, building the idea of a product to automate precisely the manual work of discovery, backed by their own research with PMs in the market, which confirmed the pain. In the session, I drew a boundary I consider important: you can automate the manual work of discovery (desk research, compiling data, transcripts), but not the discovery function itself. A PM who wants to delegate discovery to AI is still stuck in execution, just the wrong execution.
The practical corollary, which I also pass on to people I mentor: with AI lowering the cost of prototyping, if you are not embarrassed by the first version, you spent too much time planning.