How to Run Recovery Days Without Building a New Treadmill
Extra days off after heavy weeks work right up until people start chasing them. How I decide who gets one, and the rules that stop the policy from burning out the people it's meant to protect. Part 3 of 3.
An extra day off after a big week sounds simple. Run it carelessly and you've invented a game: work even harder to earn the day, show up to the day exhausted, repeat. Same burnout, nicer calendar invite.
The trap is rewarding the load
Make the day a prize for output and the smartest people on your team will optimize for the prize. They'll pick up a seventh task. They'll leave agents running overnight and review the results at 11pm. You set out to relieve the pressure and ended up putting a price on it.
So I tie the day to how much someone carried, and only loosely to how much shipped. On an AI-heavy team those usually move together. When they don't, the load decides.
How I decide who gets one
- Parallel workstreams. Three or more tasks running at once for most of a week counts as heavy, whatever shipped.
- Review burden. Approving a lot of agent-written code is high-load work, even when the reviewer's own commit count looks modest.
- Interrupts. A week of pages and incidents on top of the planned work earns a day by itself.
- The manager's read. Someone who sounds fried on Thursday is giving you data.
The manager proposes the day. Nobody should have to ask. For a lot of engineers, asking for rest feels like admitting weakness, and the ones who need it most will ask least.
Rules that keep it honest
- Use it within two weeks. Recovery doesn't keep in the freezer.
- No Slack and no "quick question." A day off with one call in it is a workday with a long lunch.
- Cover it properly. Someone else holds the pager and the agent queue. If the team can't absorb one person being out, you've found a bigger problem.
- Keep hours out of it. Nobody earns a day by logging sixty; sixty hours is a conversation.
- Include everyone carrying the load, including infrastructure and support people whose work never shows up as merged PRs.
What to watch as a manager
Pay attention to the people who never trigger a day. Some are pacing themselves well. Others have work that doesn't parallelize, and they're quietly feeling behind next to a colleague juggling five agents. Both deserve a conversation, and they're different conversations.
Watch the cancellations too. Days that keep landing next to weekends are fine; people like long weekends. Days that keep getting pushed because "this week is crazy" mean the policy only exists on paper. At that point you have a scoping problem, and no amount of time off will fix it.
Say why, out loud
When you roll this out, tell the team plainly: working this way is harder on the head than it looks, and the company knows it's getting the better end of the trade. I've seen that one sentence buy more trust than the days themselves.
Rest is the first half of a bigger change. The second half is compensation. If AI lets one developer produce what used to take a small team, the pay model should reflect that, most likely through a bonus tied to output. That's a separate essay and a much harder conversation with finance.
If your team's output jumped and you want to keep the people behind it, start with Part 1, or let's talk →