Guides

AI coding burnout: why agent work wears you out and what helps

Working with coding agents can be more tiring than writing the code yourself. You switch between sessions all day, there’s always something ready for your attention, and a break starts to feel like lost output. What helps is mostly structure: fewer sessions at once, alerts only for what blocks you, real breaks away from the agents, and a fixed end to the day.

The short answer

What developers describe

The most discussed thread on this so far is “I think I have LLM burnout” on Hacker News, with more than 300 comments. The same few causes come up again and again:

On Reddit the same theme shows up in threads about losing flow while waiting for Claude and in ADHD communities, where people describe Claude Code as both a relief and a trap: the last thing they touch before bed and the first after waking.

What burnout is, and what it isn’t

The World Health Organization describes burn-out as a syndrome that results from chronic workplace stress that hasn’t been successfully managed. It has three dimensions: exhaustion, growing mental distance from your job or cynicism about it, and reduced effectiveness at work. WHO treats it as an occupational phenomenon, not a medical condition (WHO).

A tiring week isn’t burnout. Signs that it’s more than that:

If that sounds like you for weeks at a time, talk to a doctor or a mental health professional. Habits help with stress; they don’t replace care.

Why agent work can be more tiring

Writing code yourself has a built-in rhythm: you think, type, run, fix, and the hard parts slow you down. With agents, the slow parts disappear and what’s left is supervision. You read, decide and approve, often for several sessions at once. That role is closer to a manager than a programmer, and it never quite pauses: one commenter compared it to a stone skipping across the water. Decision fatigue is real too. In a discussion of approval prompts, people who approve agent actions all day describe doing it on reflex, which is both tiring and unsafe.

What helps

Fewer sessions, one task each

Pick a number of parallel sessions you can actually follow, and drop to one for work that needs deep thought. Give each session one task and its own worktree. More on this in running several Claude Code sessions without losing track.

Quieter alerts

Get notified when a session is blocked, on a permission or a question, and let finished work wait for the end of a focus block. Turn phone pushes off outside working hours. See Claude Code notifications for how to set each up.

Breaks on a rhythm

Deciding each time whether you’ve earned a break is a decision you’ll lose. A fixed rhythm, such as 25 minutes of focus and 5 of rest, takes it off the table. Spend the break away from the screen and away from the agents; short breaks measurably reduce fatigue. See pomodoro with AI coding agents and what to do while Claude Code is working.

A fixed end to the day

Measure rest, not only output

Agents make output easy to count. Count something else too: breaks you actually took, evenings without a laptop, how often you stopped mid-review because you were tired. If those numbers fall while output rises, that’s the trade you’re making.

Where Meowdoro fits

We make Meowdoro, and this problem is why: it’s a focus timer in the MacBook notch for people who work with Claude Code and Codex. During a focus it shows which sessions are working, done or waiting for you, silently. During a break it hides them completely. Its statistics include care metrics alongside focus time: breaks taken to the end, how long agents waited for you, and focus rounds finished without peeking. It’s free before December 2026, then a one-time $20 (statistics · focus and breaks).

Questions

Are AI coding tools making developers burn out?

There isn’t solid large-scale research on it yet. There are many first-hand reports, and the causes people name, constant switching, supervision and no natural stopping points, are known sources of work stress. Treat it as a real risk that you can manage, not as a given.

Why is vibe coding so tiring?

Because most of the work is reading and deciding. You review more code per hour than you would write, and the waits break your focus into short pieces.

How many hours a day of agentic coding is healthy?

No number fits everyone. A better test is how you feel at the end of the day and after a weekend. If you need more and more rest to recover, cut the hours or the number of sessions.

Should I stop using AI agents if I feel burnt out?

Not necessarily. Changing how you use them, with fewer sessions, quieter alerts and real breaks, often helps more than quitting. If the signs last for weeks, talk to a professional.

Sources: World Health Organization on burn-out in ICD-11; Albulescu et al., PLOS ONE, 2022 on short breaks; the Hacker News and Reddit discussions linked above. This article isn’t medical advice.