Coddy found 43 percent of AI-using developers code past planned stop times

Coddy, a company that runs a free interactive coding school, surveyed 305 developers who use AI at work at least weekly and published its findings on 10 September in a report called the AI Coding Addiction Report. The New Stack wrote up the survey on 17 September, and its story is how we learned of it. The numbers below come from Coddy's own report.

What the survey measured

Coddy asked developers who use AI at least weekly about how they use it. Respondents were recruited through Prolific and CloudResearch Connect, ranged in age from 18 to 74 with an average of 38, and split 51 percent Millennial, 25 percent Gen Z, 20 percent Gen X, and 4 percent Baby Boomer. This is a self-report survey, not an observation study of what developers actually did, so the numbers describe what respondents said about themselves.

The main figures

Across every tool, 43 percent of respondents said they kept coding past a planned stop time. The rate varied by which tool they used: 62 percent for OpenAI Codex, 45 percent for Google Gemini, 40 percent for Claude Code, and 36 percent for GitHub Copilot. Coddy does not claim to know what causes that spread and neither will we, because the survey is not set up to separate the tool from the kind of person who picks it.

The report also asked which coding tool respondents found hardest to stop using when they wanted to stop. Claude Code ranked first on that question at 35 percent.

Four in five developers, 80 percent, said their AI use had felt more like a dependence than an advantage at least once. 74 percent said AI had made them more ambitious about what they take on.

What developers said they gave up

The survey asked whether respondents had delayed or skipped normal parts of a day to keep coding. 36 percent said breaks or time off, 32 percent said sleep, 28 percent said meals, 26 percent said errands and chores, and 23 percent said exercise. Coddy groups these under the idea of a cycle of rewards that keeps a developer coding, in the way social media apps do.

The consequence for a team building software

The productivity gain a manager sees from a coding agent has a second cost that most teams are not measuring. If a developer delivers 25 percent more work but the extra output came from an hour past their planned stop time, that 25 percent has a cost: a change in working hours that the tool made easier to justify. Hours taken from sleep and meals lead later to sick days, careless reviews, and staff leaving.

The Codex-to-Copilot spread of 26 points is the interesting number to think about. Coddy does not know whether the tool causes the behaviour or attracts developers already inclined to it, and neither do we. But an engineering manager who sees an increase in after-hours commits after a team standardises on one tool now has one place to check first.

Source

The New Stack: Study: Developers are addicted to AI, and managers are making it worse, 17 September 2026, by Steve Fenton, reporting on the AI Coding Addiction Report published by Coddy on 10 September 2026.