AI coding tools are making software engineers dramatically more productive, but early adopters like Django co-creator Simon Willison say the mental toll is steep and unsustainable at current pacing.
Simon Willison has been building software for over two decades. He co-created Django, one of the most widely used web frameworks in the world, and has spent years thinking about how developers work. So when he says that using AI coding agents leaves him mentally drained by 11 a.m., it is worth paying attention. Willison described the experience on a recent episode of Lenny's Podcast, explaining that he regularly fires up four autonomous AI agents in parallel to tackle separate problems simultaneously. The output is impressive. The exhaustion is something he did not anticipate.
His account cuts against the dominant narrative in the AI industry right now. Companies like OpenAI, Anthropic, and Google DeepMind are racing to build autonomous agents that handle complex tasks with minimal human input. Investors and executives frequently describe a near future where AI handles the heavy lifting and humans focus on higher-level thinking. But the reality for engineers actually using these tools today looks different. They are not stepping back. They are leaning in harder, managing multiple AI workflows at once, reviewing more generated code, and making faster decisions across more fronts. The work shifts from writing code to orchestrating and evaluating it, and that requires a different kind of mental energy.
Willison is careful to say he remains genuinely bullish on AI tools. They have accelerated his research, helped him build faster, and expanded what he can accomplish in a single morning. The problem is intensity, not capability. Running several agents in parallel means constantly context-switching, reviewing outputs, correcting errors, and deciding what to delegate next. It is cognitively demanding in a way that traditional coding, for all its frustrations, is not. As Business Insider reported, Willison noted that the fatigue has become more pronounced since November, when more capable agentic systems and open-source tools made parallel workflows practical for everyday use.
He is not alone in noticing the pattern. Engineers across the industry have described similar experiences in online forums and private conversations. There is a compulsive quality to it, Willison warned. When you know your agents can keep working while you sleep, the temptation to stay up another 30 minutes becomes hard to resist. He likened it to a new skill that workers have to develop, which is learning where their limits are in an environment where those limits have suddenly moved. That observation aligns with warnings from researchers including Gary Marcus, professor emeritus of psychology and neural science at New York University, and writers at Harvard Business Review, who have argued that AI tools risk stretching workers thinner rather than lightening their load.
What This Means for Startups and Engineering Teams
The implications for startups are significant. Founders and engineering leads are under enormous pressure to adopt AI tools and demonstrate productivity gains. The market is moving fast. GitHub reported that developers using its Copilot tool completed tasks 55 percent faster on average, and venture capitalists are pouring billions into AI-native development platforms. But speed without sustainability is a liability, especially for small teams where burnout can cripple an entire product roadmap.
Engineering managers need to think carefully about how they roll out these tools. There is a meaningful difference between using AI to eliminate tedious, repetitive work and using it to compress complex, high-judgment tasks into shorter timeframes. The first approach frees people up. The second squeezes them. Both look productive in the short term, but only one is sustainable. Teams that treat AI agents as a way to do more of everything, rather than to do the right things faster, risk repeating the same mistakes that the always-on work culture already inflicted on the tech industry over the past decade.
The broader industry conversation also deserves more scrutiny. Anthropic engineer Boris Cherny suggested earlier this year that the software engineer job title could disappear from the US workforce within months. OpenAI investor Vinod Khosla has predicted that most children today will never need to work. These are provocative claims, and they serve a purpose in shaping investor sentiment and public perception. But for the engineers actually living with these tools right now, the future feels less like liberation and more like acceleration without a speed limit.
Willison's experience is a useful corrective. He is not a skeptic or a luddite. He is an accomplished engineer who uses AI tools daily, sees their value clearly, and is honest about what they cost. The industry should take that honesty seriously. If the people building with AI are burning out faster than before, the problem is not with the engineers. It is with the expectations being layered on top of them.
