Justin Jackson has a question for software team leads: what happens when the PM starts shipping PRs?
His latest essay traces what he calls a 'Mexican standoff' unfolding across engineering teams — a phrase borrowed from Marc Andreessen — where Claude Code and similar tools have crossed a capability threshold that makes each role believe it can absorb the others. Engineers think they no longer need product managers to translate requirements. PMs think they no longer need engineers to ship. Designers are quietly learning to prompt.
The anxiety isn't theoretical. Jackson talked to founders and team leads who described PMs writing code, Heads of Product shipping pull requests without looping in developers, and companies retooling hiring away from specialists toward generalists. Kent Beck's observation — 'the value of 90% of my skills just dropped to $0' — keeps circulating in these circles because it names something teams are feeling but haven't yet figured out how to manage.
The collision point, in Jackson's framing, is the judgment layer: understanding users, setting goals, owning architecture. Every role already claims this as its core territory. When AI compresses the execution gap between disciplines, everyone rushes toward the same narrow band of high-leverage work — and the turf war begins quietly, before anyone has named it.
Jackson isn't entirely pessimistic. He points to Matt Stauffer, CEO of Tighten, as an early model of what comes after: a business development manager and an engineer prompting an LLM together in real time, each contributing domain expertise the other lacks. Jackson calls this cross-role AI pair programming — the PM owns product goals and user understanding, the engineer owns architecture and code quality, and both co-own the pull request. It's the 37signals two-person team model, updated for the LLM era.
Whether that collaborative structure takes hold or gives way to zero-sum competition depends on whether teams can develop new norms before resentment sets in. John O'Nolan of Ghost thinks new roles will emerge from the compression. Jackson seems to hope so, though his piece doesn't hide that the disruption is happening now, not in some hypothetical future.
The stakes for anyone building in the agent space are clear enough: as tools like Claude Code lower the implementation barrier, execution alone stops being a competitive moat. The teams that learn to pool judgment across disciplines — not just hand off tasks between them — may have a structural advantage that compounds over time.