Midjourney founder David Holz just put a name on something every developer using AI coding tools has felt this year: the strange combination of superhuman output and deep exhaustion. In a post on X, Holz said his friends are "all feeling extremely productive and also extremely drained with the latest coding models." His next line is the one getting the most attention: "This makes me feel like something is wrong, and also that there might be a big opportunity."
Holz ended the post with a question that turned the thread into an informal group session for the AI era: "Does anyone have any strategies they use to make it feel better day-to-day?" The answers that followed are worth saving, because they come from people who live inside this world every day: engineers, product leaders, and designers who have worked at Anthropic, Meta, X, and Cash App.
The timing matters. Holz is not an outsider watching the coding-agent boom from a distance. He runs one of the most successful AI startups of the decade, and his own company now competes for the same developers who are feeling the drain. When he says something feels wrong, the whole industry tends to listen.
What the Comments Revealed About AI Coding Fatigue
The thread quickly split into two camps: people diagnosing the problem, and people offering fixes. Shuming Hu, a former Meta engineer, put a sharp label on it: "vibe coding doesn't get you into a flow state." That one sentence explains a lot of the drain. Prompting a model to generate code feels productive in the moment, but it skips the deep engagement that makes programming satisfying in the first place.
Catherine Wu, head of product for Claude Code at Anthropic, offered the opposite strategy: "focused work on a hard task with a single agent." She admitted she usually runs "10s of agents" at once, but finds it better to "get into the zone and in the weeds of one task to get it done well."
Brandon Kainoa Jacoby, a former designer at X and Cash App, warned the feeling is "probably going to get worse before it gets better." His practical tip: do a deep cognitive task entirely away from any model. Other commenters went even more analog, suggesting walks, staring at a tree, or playing with their kids.
The Fatigue Behind the Headlines
This is not a one-off complaint. In February, programmer Siddhant Khare published a viral essay arguing that AI fatigue is real and "nobody talks about it." Developers have told Business Insider they struggle to keep up with the pace of change, describing a kind of workplace paralysis. The pressure is compounded by the fear of falling behind: Ben South, a serial founder and former Postmates VP, wrote in Holz's thread that "even an hour of rest feels like a ton of productivity lost."
That sentence captures the trap. When rest feels like a loss, the drain becomes a treadmill. The tools get faster, expectations rise, and the human at the keyboard never gets a break. Holz's instinct that "there might be a big opportunity" is telling: the company that figures out how to make AI-assisted work sustainable could own the next era of developer tools.
5 Ways to Use AI Coding Tools Without Burning Out
Until that company exists, you can steal the strategies that actually showed up in Holz's thread. Here is a practical checklist:
- Run one agent, not ten. When the task matters, close the extra sessions and focus on a single agent working through one hard problem. Deep focus beats parallel busywork.
- Protect a no-AI window every day. Spend 30 to 60 minutes on a hard cognitive task with no model in the loop. It rebuilds flow and reminds your brain it can still do the work.
- Treat rest as part of the job. If an hour of rest feels like lost productivity, your schedule is broken, not your discipline. Block real breaks and defend them like deadlines.
- Watch for vibe-coding drift. If you are generating code faster but understanding it less, slow down and read what the model wrote. Flow state is where the satisfaction and the learning happen.
- Go analog when the screen gets loud. A walk, a tree, a conversation with a human. The most popular advice in the thread was also the simplest.
Holz is right that the current state of AI coding feels like something is wrong. But this is also the first time in a long while that the industry is openly discussing the human cost of its own speed. The developers who learn to manage the drain, not just the productivity, will be the ones still standing when the next model drops.
Comments