In partnership with
Premium membership for The Modern Leader now lives in one place. Take a look and find the level that fits: full archive access, Leadership Hotline answers, templates and tools, plus courses and direct time with me at the higher tiers.
My workday doesn't really end anymore, and I don't think I'm alone in feeling like this.
I keep hearing people joke about giving an agent "just one more nudge" before logging off, or about checking in on one while they're out for a run. Isn't running hard enough? Do we really need to be thinking about what our agent needs from us as we run up a long hill?
It's easy to see why people do it. AI lets you keep several things moving at once: while one agent runs, you can review another's output, send a follow-up between meetings, or check back that evening to see what finished. Before agents, the day had its own built-in stopping points, too: you finished the task, you were waiting on someone else, you hit a problem that needed fresh eyes in the morning. Those pauses don't really exist anymore. There's always one more thread to pull instead.
I wrote about an early version of this problem almost a year ago: AI promised to make us more efficient, and the time we saved just got filled with more work. Agents have only made that worse. The agent doesn't get tired, but you do.
A thank you from this week’s sponsor:
Your data exists. Your team just never sees it.
Your numbers live in dashboards nobody opens. SureThing pulls them from the tools you already run and drops one plain-English report in Slack every morning.
In its 2026 tech workforce survey, Lenny's Newsletter found that significant burnout jumped from 44.7% to 55.7% in a year, while career optimism fell from 54.8% to 48.7%. The people who felt most destabilized by AI were also the most burned out and the least optimistic.
This has a business cost too, separate from the empathy argument. A rushed review of an agent's output at 9pm misses things a focused pass at 10am would catch, and the rework that follows usually costs more than what the agent saved in the first place. The engineers who can run five threads at once, typically your most senior people, are also the ones burning out fastest trying to keep up with that volume. Agents made throughput look infinite. Attention never did, and that's where the real bottleneck lives now.
AI isn't responsible for every layoff or reorg shaking the industry, but it's changing our relationship to work. When useful work is always available, the workday can't tell you when it's over. You have to decide that yourself. That's simple to say and hard to do: it takes real discipline to leave an agent running through lunch, let finished output sit until tomorrow, or close the laptop knowing there's more you could squeeze out. The cost of checking in always feels low. But those small check-ins add up, and they're what keep you half-tethered to work during the hours you're supposed to be resting.
That's a lot to ask of any one person to manage alone, which is why it can't just be an individual habit. A workflow that can run indefinitely needs someone deciding when to shut it down, and that has to be built into how the team operates, not left to individual willpower.
Start with the incentives that are actually in place, not the ones you intend. If someone merges a PR an agent finished at 11pm and gets a "nice work" in the team channel the next morning, that's the signal that sticks, whatever your stated policy says. A better norm: agent output waits for the next business day unless something is genuinely blocking someone, and that gets said out loud in standup, not buried in a doc nobody opens.
This falls apart if managers don't hold themselves to the same standard. If you're the one pinging an agent from the couch at 9pm, or reviewing a PR while you're supposed to be on vacation, that's the behavior your team will copy, no matter what you say in the 1:1.
None of this matters much if the work being run through all those agents isn't worth doing in the first place. Output and impact aren't the same thing, and it's easy to lose sight of that difference when agents make it cheap to produce more of both. If AI adoption is the mandate at your company, the goal isn't to log more hours working, it's to get more of the right things done. That means going to your PM and ruthlessly cutting the backlog down to what's actually worth doing, instead of letting agents plow through a queue that was already full of busy work.
The same thinking applies to time away, not just daily habits. Track who hasn't taken a real break lately, and ask about their next vacation, not a long weekend but a full week where their brain actually stops cataloging what's waiting when they're back. That means signing out of Slack, maybe deleting the app for a week, and resisting the urge to peek at what the agents produced.
PTO won't fix a workload that's fundamentally too much. People come back to the same pace, so leaders still need to watch whether that pace is sustainable. But that's a reason to keep managing the load, not a reason to skip the time off. I'll be writing more about taking real time off next week.
AI has erased most of the workday's natural stopping points. We can't wait for the work to be finished before we rest, because it may never be. Sometimes the only option is to decide you're done, close the laptop, and let the agents keep running without you.