AI saves developers 13 hours a week. None of it came back.

The pitch for AI coding tools was time. The data says the time showed up and then went somewhere else. BairesDev's Q3 2026 Dev Barometer reports developers saving 13 hours a week on coding — up from about seven a year ago — and not one of those hours turned into slack. They moved a layer up, into reviewing, debugging and securing machine-written code. If you are paying for AI coding seats and expecting a straight-line throughput win, this is the number that explains why you did not get one.

What actually happened

BairesDev published the Q3 2026 Dev Barometer on September 15, surveying 705 developers across 60+ countries plus 41 enterprise CTOs. The headline shift: the share of developers saying AI writes at least half their code went from 12% in Q3 2025 to 42% now. Nearly 80% now spend less than half their week writing new code.

The interesting half is where the hours went. 67% report reviewing more AI-generated code than a year ago. 52% spend more time debugging problems AI introduced. On the buy side, 78% of CTOs increased spending on code review, QA and validation. That is not a side effect — it is the same money, relocated.

Two more numbers worth keeping. The skills developers say drive raises are AI tool fluency (29%), system architecture (20%), and communication (15%) — but only 25% of CTOs actively invest in that last category. And 86% of developers find the role more fulfilling, up from 76%. This is a survey of self-reported time, not instrumented telemetry; treat the direction as solid and the decimals as soft.

Why it matters for your business

If you run a small team or hire a shop, the practical read is this: AI moved your bottleneck, it did not remove it. Generation is no longer the scarce thing. Judgment about generated code is. That is a harder thing to buy, and it does not come with a per-seat price.

Three things follow.

Your cost model is wrong if it only counts seats. A $39/month license that produces three times the diff volume creates review load that lands on your most expensive person. The 78% of CTOs raising QA spend are not being cautious — they are paying the bill that arrived with the tool. Budget for it deliberately instead of discovering it as senior-engineer burnout.

Review has to be a real step, not a rubber stamp. Volume kills attention. Concretely: keep PRs small enough that a human can actually hold them, require tests that a human wrote or at least read, and put automated gates — type checks, lint, dependency audit, a security scan — in front of the human so the human spends their attention on logic, not formatting.

And measure the thing you care about. Not lines shipped, not PRs opened. Time from problem reported to fix live, and how often a change has to be reverted. Those two move in the right direction only if the review layer is real.

Sources: BairesDev / GlobeNewswire, VentureBeat.