A year ago, half of developers were willing to let AI write their tests. Now three quarters are. The bigger jump is in application logic, the core of most software, where readiness went from 18% to 50%. Developer AI delegation is moving fast, but it isn't moving evenly, and the parts that stay with people say a lot about how software teams will work next year.
The numbers come from JetBrains Research, published this month, which compared its 2025 and 2026 developer surveys. Willingness to hand refactoring to AI rose from 38% to 60%. Two activities held out. Only 43% would fully delegate code review, up from 33%, and 54% still aren't ready to delegate debugging, down from 76% a year ago. Meanwhile 79% expect employers to require proficiency with AI tools.
Where developer AI delegation stops
The line falls in a sensible place. Writing code is easier to check than to produce, so handing it off is a reasonable trade. Review and debugging are different. They ask whether the code does what the business actually needed, and why something broke in a system with history. Those are judgment calls that depend on context the code itself doesn't hold.
The Stack Overflow 2026 Developer Survey points at the same gap from another side. When developers named the context AI agents most need, 85% said project goals or requirements and 73% said code and documentation. Only 44% listed prior decisions and rationale, even though that's often exactly what a reviewer or debugger reaches for. The survey's summary also notes that 73% of people using AI coding tools use them daily, so this is everyday practice now, not an experiment.
The practical takeaway is less about which tasks to give the tools and more about what to write down. If people keep the review and the debugging, they need the intent behind a change, the reasons for past decisions and a clear owner for each part of the system. Teams that keep that record get more out of AI-written code, because a human can still tell whether it's right.