Staff the line: spec author, verifier, fleet ops, line owner
Augment’s org-design guide: agents absorb implementation, so scarce human work is spec, verification, halt authority, and line ownership — with receipts on review share collapsing under PR flood.
The useful move in Augment’s org-design guide is composition, not headcount. Agents take implementation; the scarce human jobs become four named accountabilities on a delivery line: spec author (outcome, bounds, acceptance as a versioned artifact), verifier (scenarios, scored validation, which findings gate merge), fleet operator (environments, experts, triggers — and the person who can stop every agent at once), and line owner (merge and halt authority on the same person, written down so nobody has to approve the stop).
Receipts, not slogans: one enterprise arXiv panel (802 developers, ~196k PRs, Jan 2024–Apr 2026) under a double-throughput mandate saw raw PR volume 3.1× while the reviewer pool grew only 1.5×, and the share of PRs with at least one human review fell from 89% to 68%. Bun’s Zig→Rust rewrite (via PE) is cited as ~15% implementation / ~85% fix-test-verify. LeadDev/Gartner-side notes on write access without revoke paths, and PagerDuty’s point that service ownership no longer follows authorship, land the operational cap: on-call load limits how many agents one line owner can carry.
Why it matters for code review
Review capacity is an org-design problem before it is a bot problem. If halt authority is decorative, seniors are still a queue, and juniors lose the path that taught the system, you did not staff a factory — you staffed a flood. Name the verifier and the line owner in writing, drill the halt, and size span of control by evidence the owner can actually read.