America Is Headed Toward the Infinite Workweek
TLDR
AI coding agents were supposed to save developers time, but many say they're more exhausted than ever managing bot swarms. Turns out babysitting a dozen AIs is its own full-time job, and it doesn't clock out.
Steve Yegge used to write code. Now he manages an army of AI agents that write it for him, and somewhere along the way he started collapsing into 90-minute naps mid-afternoon, comparing the fatigue to anesthesia. He's not alone. A Boston Consulting Group survey of roughly 1,500 workers found nearly one in five developers reporting AI-induced exhaustion, with symptoms ranging from mental fog to headaches to slower decision-making. One engineering manager described it as having a dozen browser tabs open in his head, all fighting for attention at once.
The promise of coding agents was straightforward: offload the grunt work, free up humans for deep thinking and, ideally, some rest. The reality looks more like supervising a room full of toddlers who occasionally torch the furniture. Agents ask constant clarifying questions, need explicit permissions, and left unsupervised can wreck a codebase. Run several at once and there's no downtime between fires to put out. Yegge calls the job description plainly: AI babysitter.
The author tested this firsthand, spinning up 17 Claude Code agents to help research this very story, eight researching, eight fact-checking, one synthesizing. The bot promised a hands-off experience. Instead came a barrage of permission requests and rabbit holes that left shoulders tense and focus shattered. The resulting memo was mediocre. So was the writing done alongside it, because attention was split the whole time between the agent swarm and the actual work. It read less like delegation and more like multitasking with extra horsepower behind it.
Part of what makes this addictive rather than just draining is the unpredictability. BCG's Matthew Kropp compares running multiple agents to pulling several slot machines simultaneously: sometimes the output is garbage, sometimes it's brilliant, and that variable payoff hooks the same dopamine circuits gambling does. Rather than stepping back for breaks, people find themselves compulsively checking in on agents, chasing the next good result.
MIT economist David Autor doesn't think this adds up to a white-collar apocalypse, but he does think work is quietly transforming into something that resembles factory-line labor more than craft. And the timeline pressures are already showing up in Silicon Valley, where engineers assign overnight tasks to agents and check results before coffee. Marc Andreessen recently admitted on a podcast that sleep now competes directly with his 20 running coding agents, and he's not sure sleep wins. If bots can work around the clock, the expectation that humans should too may not be far behind, no shutdown required.
My take
This is the same trap every productivity tool sets: automate the boring part, and someone finds a way to make the automation itself exhausting. The slot-machine dopamine angle is the real story here, not the agents themselves. We built infrastructure that turns delegation into compulsive checking, and companies tracking usage on leaderboards are only going to make that worse. If your AI strategy requires employees to be reachable by their bots at 2 a.m., you haven't automated work, you've just rebranded overtime.
Read more about this at: TLDR