AI Now Writes Half of Code for 42% of Developers, But Saved Hours Go to Review, Not Rest

The thesis is straightforward: AI is generating more code than ever, but it is not generating more free time for the people who write software.

BairesDev's Q3 2026 Dev Barometer, a survey of 705 developers across more than 60 countries plus 41 enterprise CTOs, found that 42% of developers now say AI writes at least half their code. A year earlier, in Q3 2025, that figure stood at 12%. That is a 3.5x increase in twelve months, the kind of adoption curve that would normally signal a labor market disruption.

Time saved has grown too. Developers report AI now saves them 13 hours of coding per week, nearly double the roughly seven hours reported a year ago. On paper, that looks like a windfall of spare capacity.

It isn't. Only 21% of developers now spend more than half their workweek writing new code from scratch. Meanwhile 67% say they spend more time reviewing AI-generated code than a year ago, and 52% report spending more time debugging problems the AI introduced. Time spent learning AI tools and other new technologies has more than doubled, from roughly four hours a week to nine.

BairesDev CEO Darren Shimkus put it bluntly in the release: "Developers nearly doubled the time AI saves them in coding, to 13 hours a week. Not one of those hours came back. A year ago we read the first seven hours as capacity, and we got that wrong."

Why It Matters

This is a labor market story before it is a technology story. If AI coding tools simply compressed engineering hours, the implication would be headcount reduction. Instead, the BairesDev data suggests the bottleneck in software development is migrating from code generation to code validation, and that shift changes what an engineering organization needs to staff for.

The pattern lines up with independent research. Stack Overflow's 2025 Developer Survey, with more than 49,000 responses, found 80% of developers using AI tools in their workflows, but only 29% trusted the accuracy of the output. Sixty-six percent said they were spending more time fixing AI code that was "almost right." Three-quarters said they would still turn to another human when they didn't trust an AI-generated answer. Adoption has outrun trust, and that gap is exactly where the extra review and debugging hours are landing.

JetBrains' 2026 Developer Ecosystem Survey, based on more than 15,000 professional developers, adds another data point: 90% were using AI coding agents at work at least weekly between May and July 2026, with 68% using them daily. Agents are no longer a novelty. They are infrastructure. But infrastructure needs governance, and that is where enterprises appear to be spending.

Among the 41 CTOs BairesDev surveyed, 78% said they had increased spending on code review, quality assurance and validation specifically to support AI-generated work. Only 7% of developers said the decision to ship code had been delegated entirely to AI without human input. That is a telling number. It says most organizations still treat AI output as a draft, not a deliverable.

Shimkus told VentureBeat he expects developer accountability for code to move toward 100%, even as the share actually written by AI keeps climbing. Sophisticated enterprise customers, he said, increasingly won't let an engineer check in code they can't explain. "In a world where we don't have AI explainability, I need developer explainability," he said. He drew a clear line between AI-generated prototypes, where companies tolerate far more autonomy, and production software serving thousands of companies or millions of consumers. That code, he said, "will not happen without review, at least not anytime soon."

There's a methodological caveat worth flagging. The 705 developers surveyed are not BairesDev's own engineering staff. The company says most are applicants going through its screening process, not employees. The CTO sample of 41 is also small relative to the developer sample. That doesn't invalidate the trend, but it should temper how precisely these percentages get treated.

The shift hasn't dented morale. Eighty-six percent of developers now say AI makes their role more fulfilling, up from 76% in Q3 2025. Shimkus compares the change to what computer-aided design did to architects: CAD automated manual drafting without eliminating the profession, freeing architects to spend more time on design. "The job hasn't gotten smaller," he told VentureBeat. "It's moving up a level." For engineers, that higher level increasingly means architecture, problem definition, review, security and understanding how generated code interacts with larger systems.

The trajectory was visible earlier. BairesDev's inaugural Q3 2025 Dev Barometer found developers saving 7.3 hours a week on coding, with 76% reporting more fulfilling work. By Q4 2025, 65% of developers expected their role to be redefined in 2026, and among that group 74% expected to spend less time coding and more time designing solutions, with half foreseeing greater emphasis on architecture and strategy.

Compensation is starting to reflect the reshuffle. Among developers who received raises in 2026, 29% attributed it to AI tool fluency, 20% to system design and architecture, and 15% to human skills like communication and cross-functional collaboration. Yet only one in four CTOs said they were actively investing in developing those human skills, a mismatch worth watching.

Shimkus's advice to developers is to go deep rather than wide: master one AI tool rather than sampling many, and prioritize security over incremental coding fluency. "If I had to advise anybody, should you go even deeper in your code development infrastructure, or to go deeper in security, go deeper in security," he said. That advice tracks with separate research reported by ITPro on agentic software engineering, which found a widening gap between agent adoption and the governance controls, including deployment gates, that organizations have put around them.

The Bottom Line

AI is not shrinking the engineering workforce's workload; it is redistributing it toward review, security and judgment. Generating code is getting cheap. Deciding whether that code belongs in production remains expensive, and increasingly that expertise, not raw coding speed, is what compensation trends and hiring priorities are starting to reward.