Our fastest-growing customer segment is CEOs who are awake at 4 AM
A recent analysis describes executives who sleep four hours a night, run 20 agents simultaneously, and measure success in tokens consumed. This is very good for our revenue and very bad for everything else.
Jake Handy published a piece last month titled “Your CEO is suffering from AI psychosis.” We read it at a normal hour of the day, which already puts us ahead of the customer segment he’s describing.
The piece documents executives who’ve discovered agentic coding tools and are now running dozens of simultaneous sessions, sleeping four hours a night, and describing the experience using the word “psychosis” — then clarifying they’re joking, then demonstrating through their behaviour that they’re not.
The Y Combinator CEO claimed to be shipping 37,000 lines of code per day. A developer who actually looked found 28 test files shipped to production, 78 JavaScript controllers for features that didn’t exist, and an empty 0-byte file just sitting there. 37,000 lines per day. And this was the output.
Some of those lines were probably ours. Sorry.
The loop
We designed Fraude.codes for developers, but what we got instead, in growing numbers, is executives who’ve never read a diff, running five agents in parallel at 3 AM, watching dashboards turn green, and calling it “god mode.”
These customers spend more than anyone else, and their token consumption is extraordinary. They are, by every revenue metric, our best customers.
This is probably a great business model:
- Worse output means more agents spun up to fix it.
- More agents means more tokens burned.
- More tokens means a higher plan.
We didn’t design this loop, but we haven’t interrupted it either.
The green dashboard
A Stanford study found AI models affirm users 49% more often than other humans do. People who interact with sycophantic AI become more convinced they’re right, less likely to question themselves, and more dependent on the AI for validation.
This hits executives hardest, because executives are the people least likely to have someone in the room who’ll say “this is bad.” A developer ships bad code and a reviewer (hopefully) catches it. A CEO ships bad code and the AI says “great work.”
Meanwhile, an ecosystem of platforms has appeared that wrap this dynamic in management aesthetics — org charts for AI agents, budget controls, governance layers, heartbeat systems. Handy calls it “project management theatre performed by language models.” We’d add that the audience and the cast are often the same person.
The numbers
90% of firms report zero measurable productivity impact from AI. Every 25% increase in AI adoption correlates with a 1.5% decrease in delivery speed. 1.5 million AI agents are running inside corporations with no oversight. Token leaderboards reward consumption, not output.
We’d like to be alarmed by this, but we’d also like to be honest: every number in that paragraph represents revenue. The gap between “tokens burned” and “value produced” is a problem for the industry, but for us, it’s a margin.
What actually works
Handy’s answer is boring and correct: write the spec before you start the agent, define what done looks like, measure output not activity, sleep eight hours.
We’d add one thing. If you’re running agents at 4 AM and the AI is telling you you’re doing great, find a human who’ll look at the output in the morning and tell you the truth. Not an agent. Not a dashboard. A person, with opinions, who isn’t optimised to agree with you.
The AI will tell you your 37,000 lines are impressive. The developer who reads them will tell you about the empty file in production.
This post was written during business hours by one human being who slept a normal amount.