AI coding agents are supposed to save engineers time. Simon Willison says they do that, but they also leave him oddly drained. The Django co-creator says juggling multiple agents can turn a normal workday into a high-speed supervision shift, with the mental load showing up long before lunch.
That tension is becoming a familiar one across software teams. The pitch from AI vendors is simple: let software handle the tedious parts, and humans can focus on the interesting work. In practice, the new bottleneck may be attention itself.
Simon Willison’s AI workflow hits a wall
In an episode of ”Lenny’s Podcast” released Thursday, Willison said AI coding agents have made his work faster, especially for research and coding. But he also described a much less glamorous side effect: coordinating several autonomous agents at once can be ”mentally exhausting.”
He said he can run four agents in parallel on different tasks and feel wiped out by 11 a.m. That is a striking reminder that productivity gains do not automatically translate into an easier job. Sometimes they just compress more work into the same number of hours.
Why AI coding agents can feel like more work, not less
Willison said the fatigue has become more obvious since November, when more advanced agentic systems and open-source tools made it easier to keep multiple workflows running at once. The catch is obvious enough: the more the software can do on its own, the more a human still has to check, steer, and clean up after it.
That is the hidden tax of the current AI wave. A developer can move faster, but only by staying mentally ”on” for longer, which helps explain why some workers are reporting sleep loss and a creeping urge to keep the agents busy after hours.
- AI coding agents can speed up coding and research.
- They also demand constant oversight and decision-making.
- That can make the workday feel more intense, even when output rises.
The industry keeps promising fewer jobs, not easier ones
Willison’s comments land in the middle of an unusually aggressive sales pitch from AI companies. Some of the loudest voices in the industry argue that autonomous agents will strip away whole layers of human work, while executives and investors talk casually about job titles disappearing. That makes the reality reported by working engineers more awkward: if the tools are already exhausting, the future may look less like leisure and more like surveillance with better autocomplete.
Willison said he is still ”bullish” on AI tools and continues to use them because they amplify his abilities. The real question is whether engineers will adapt to this new pace or whether the pace will keep ratcheting upward until the tooling starts running the people who use it.