AI shows up at night

Writing · 25 Oct 2026

AI shows up at night

What a summer of coding with AI did to my bedtime

Since this summer I build software together with AI agents, and I noticed something: I keep going later into the evening. But not against my will! Every evening there is a new possibility to experiment with, and I want to see what is possible before I close the lid. The question is whether that feeling is real, or just a gut feeling. I have worn a WHOOP band since October 2019, so I have seven years of bedtimes, wake-ups and sleep scores. I compared the months since June 2026 with the same months of 2024 and 2025, work nights only, Monday to Friday. The feeling is real. VERY real.

Median bedtime per month. Blue: 2024 and 2025. Orange: 2026, from June the months with AI agents on the desk
Median bedtime per month. Blue: 2024 and 2025. Orange: 2026, from June the months with AI agents on the desk

The shift starts in June 2026. After a year of careful, partial experiments, from vibe coding to rapid iteration in agentic loops, I decided at the end of May to go 300 % in: full throttle on AI and agents like never before! Six websites in ten days was the first result, and the pace has been twenty times faster ever since! The bedtime follows that decision and holds every month: up to an hour and twenty minutes later than the same month a year earlier. July is the clearest, 23:32 against 22:14 in 2024 and 22:13 in 2025. The early bedtimes are gone: a third of my work-night sleeps used to begin before 21:30, now it is one in nine. And I get up twenty minutes earlier as well, so the sleep is squeezed from both ends. An hour a night, on average: that is what my enthusiasm costs!

Work nights, April to October 2024 2025 2026
Median bedtime 22:05 22:01 22:56
In bed after 23:00 27 % 22 % 49 %
In bed after midnight 12 % 13 % 25 %
In bed before 21:30 36 % 31 % 11 %
Wake-up 06:43 06:45 06:24
Hours asleep 7.6 7.9 7.2
6 July 2026, 00:13. Who needs sleep if you can build amazing things?!
6 July 2026, 00:13. Who needs sleep if you can build amazing things?!
Work nights, April to October: the early nights disappeared and the late ones doubled
Work nights, April to October: the early nights disappeared and the late ones doubled

Conclusion

What it has not cost, so far, is my body. My heart rate variability and recovery scores for June to October are as good as last year's, better even, and my resting heart rate is the lowest since 2020, to which the running probably makes a big contribution. Still, I keep a close eye on the data and keep listening to my body. The enthusiasm and my optimism are big; I just have to make sure the (health) balance does not tip. THAT is what I watch!

26 August 2026, 23:06. When the monitor is too small, the television will do
26 August 2026, 23:06. When the monitor is too small, the television will do

← More writing