AI agents push founders to 20-hour days as burnout deepens

AI agents were supposed to give founders back their time; instead they are consuming it, and the sleep that goes with it.

AI agents built to automate routine work are keeping startup founders tethered to their screens around the clock, with some logging 20-hour days to supervise bots that never sleep.

"The cost of the agents' being blocked for eight hours is way too high," said Aditya Sharma, 27, co-founder of Keel, an AI troubleshooting platform for IT teams. "They can be done with their work at any point of time, in the middle of the night."

Sharma recently went to bed at 6 a.m. after an all-nighter with his crew of generative-AI bots, which build customer features, conduct research for an in-house model and monitor marketing campaigns. Peter Pezaris, 56, who launched his fifth startup, Proxon, four months ago, typically works from 7:30 a.m. to 2 a.m. — a schedule he describes as "like a drug." Proxon, which employs six human developers, is operating 30 times faster than it would without agents, he estimates, with an annual revenue run rate expected to exceed $1 million by October.

The pattern cuts against the productivity narrative driving AI adoption across Silicon Valley. A National Bureau of Economic Research working paper surveying nearly 6,000 CEOs and senior executives found more than 90 percent reported no effect on employment from AI, and 89 percent saw no change in labor productivity. Yet founders say they cannot stop — the fear of missing a week's worth of agent work in a single idle hour keeps them checking in.

The irony is structural. AI agents — systems that plan and execute multistep tasks like pulling data and writing code — require human guidance and context as they move through their work. That means founders become round-the-clock supervisors, monitoring agents that can finish tasks at any hour and stall indefinitely without input.

Ajay Kalia, 43, a Boston-based founder who left Spotify's AI group in 2024 to launch Alt, starts most days at 5:30 a.m. and wraps up around 9 or 10 p.m. He recently began controlling agents from his Apple Watch, which buzzes every 10 minutes with another agent request. "I don't know if it's healthy for me to be out on a run looking at my watch to give my agent permission to do something," he said.

The workload compounds. Faster onboarding at Proxon means more customer requests to handle. Riveter, an AI-enabled data-gathering platform run by just two people — co-founder Abby Grills, 33, and her partner Cody Watters — keeps deferring hiring because agents keep absorbing new tasks. Grills works 10-hour days with one weekend day off. "We think we can do it, just the two of us," she said.

The burnout risk is not hypothetical. Pezaris was hospitalized twice during earlier startup stretches, once sleeping an hour a night for a month before collapsing at the office. Grills watched a fellow Y Combinator founder in her 2024 cohort end up hospitalized after living on meal-replacement products. Y Combinator says it has long advocated that founders take care of themselves.

The productivity paradox

The disconnect between AI's promise and its measured impact is widening. PwC's 2026 Global CEO Survey of 4,454 chief executives across 95 countries found 56 percent had seen neither revenue nor cost benefits from AI, while only 12 percent reported both. A separate survey of 5,000 white-collar workers found 40 percent of non-managers said AI saved them no time each week, while 19 percent of executives said it saved them more than 12 hours.

For founders building on AI, the calculus is different. The technology is not just a tool — it is the product, the workforce and the competitive edge. Kalia said every few months he has to recalibrate his work around rapidly evolving model capabilities. "You can build these bigger and wildly more ambitious concepts with these new models, which is fantastic," he said, "except you're never done."

The stakes extend beyond individual founders. If the pattern holds, the AI boom could reproduce the worst excesses of earlier tech cycles — the punishing schedules, the hospitalizations, the burnout-driven churn — at a moment when the industry is already struggling to show productivity gains in aggregate data. Forrester Research projects that half of AI-attributed layoffs will be quietly reversed as companies discover they cut before redesigning work.

For investors, the signal is mixed. Companies like OpenAI, Salesforce and Anthropic are pouring billions into agentic AI infrastructure, betting that autonomous agents will eventually deliver the productivity gains that have so far failed to materialize. But the founder experience suggests the human cost of supervising these systems may be higher than the productivity math implies — and that the real bottleneck in AI adoption may not be model capability, but human capacity.

This article is for informational purposes only and does not constitute investment advice.