Developers Are Getting Addicted to AI Coding Agents, Tool Makers Warn

A growing number of programmers admit that agentic coding tools are literally frying their brains. According to an Axios report, the most popular systems, such as Anthropic's Claude Code, OpenAI's Codex, and the open-source OpenClaw, are triggering behavior in some of the most productive software builders that looks strikingly similar to gambling addiction.

The mechanism described by Axios's sources resembles the design of slot machines. The agent receives an instruction, returns a result a moment later, sometimes a success, sometimes a failure, and that unpredictability fuels the same dopamine cycle known from gambling.

They work like slot machines. You type one prompt, get a response, a piece of code is ready - Quentin Rousseau, co-founder and CTO, Rootly

The variable reinforcement mechanism

In an earlier blog post from March 9, 2026, Rousseau cites an observation by engineer Steve Yegge, according to which agentic coding operates on the principle of variable reinforcement known from the psychology of gambling. Sometimes tests pass flawlessly, sometimes they fail spectacularly, and that uncertainty produces alternating spikes of dopamine and adrenaline.

It is this pattern, not simply the volume of work, that sets the new phenomenon apart from classic overwork. Rousseau lists traits specific to agentic coding: the absence of cognitive friction, since a prompt replaces active thinking, a spectator effect from watching the agent work, tasks branching endlessly, and the social gamification of results on social media.

My brain just got stuck in go mode - Quentin Rousseau, co-founder and CTO, Rootly

Overwork culture as achievement

Rather than warning against this pattern, the startup world often treats sleep deprivation as proof of productivity. Y Combinator chief Garry Tan publicly described spending 19 hours without sleep working with Claude Code, calling the experience addictive outright, though he later admitted it wasn't healthy.

Rousseau warns that the first casualties of this trend are the people most invested in deploying AI, namely founders and engineers working at the front line of innovation. According to him, peak productivity requires working with as many agents as possible at once, which means near-constant context switching and mounting strain on a limited attention bandwidth.

The scale of the phenomenon

Axios's report also draws on broader research into AI-related cognitive fatigue. Boston Consulting Group, together with researchers at the University of California, Riverside, are studying a phenomenon known as 'brain fry,' and Harvard University surveys point to mental fog reported by employees who use AI tools intensively. There are also reports in the background of higher staff turnover, decision fatigue, and elevated error rates among those most hooked on agentic tools.

What it means for developers

For Polish development teams increasingly reaching for Claude Code, Codex, or similar tools, the report is a signal to treat agentic coding as work that requires breaks rather than constant real-time monitoring of results. Rousseau himself admits that his attempts to curb the habit have had limited success, but he recommends running agents in batches instead of watching them work live, tracking hours worked rather than just the volume of code generated, and having open conversations about compulsive behavior instead of celebrating it.

The phenomenon is gaining weight at a time when tech companies increasingly measure team performance by the number of agents launched and the pace of code delivery. Pressure to show off results on social media further rewards extreme work marathons, which, according to the people cited in the report, makes it harder to treat the problem as a genuine health risk rather than a curiosity out of Silicon Valley.