Agent addiction: 80% of developers report dependence on AI coding tools over advantage

Quentin Rousseau, CTO and co-founder of Rootly, posted on LinkedIn that he found himself at 02:47 watching Claude Code refactor a module with no deadline and no bugs — and simply couldn't stop. He described agentic coding as a dopamine hit when the agent succeeds and an adrenaline spike when it fails, a cycle that cost him sleep and eventually drove him to seek medical help. His story is not an outlier: a survey by training company Coddy Tech of 305 developers found that four in five (80%) feel their use of AI tools resembles dependence more than advantage.

The data reveal a clear paradox. 74% of respondents said heavy tool use had increased their chances of a raise or promotion, yet 51% admitted they were on a path to burnout. More than two-fifths (43%) keep coding with AI late into the night after intending to stop, 32% delayed sleep to continue, and 39% said the tools make it harder to disconnect at day's end. Watching the agent work feels passive enough to pass for rest, yet active enough to keep the developer in the loop.

The picture sharpens in Stack Overflow's 2025 developer survey. Tool adoption climbed to 80% of respondents, but trust in accuracy fell from 40% in prior years to 29% this year, and overall sentiment dropped from 72% to 60%. 45% of developers described frustration with answers that are "almost right, but not quite" — output that looks convincing but creates difficult debugging work. The gap between ease of generation and complexity of verification creates what is being called "verification debt": code arrives fast, but the developer stays stuck checking whether it is correct, secure, maintainable, and fit for the specific architecture.

The problem compounds when leadership treats AI output as a pure force multiplier. If an organization views the tools as a way to double throughput without pricing in the costs of review, security, and integration, the dependence Coddy documented becomes a systemic trap. Developers are still required to understand requirements, spot architectural conflicts, cover edge cases, and own production — all while the agent keeps generating more code that needs checking. The result is not higher productivity but an endless, wearing loop of production and verification.