More people are handing long jobs to AI agents.

Refactor this whole codebase. Process these hundreds of files. Run the tests and fix what fails. Go research this and report back in thirty minutes. While the agent works, you do something else — or you walk away.

And thirty minutes later you come back to a stopped job.

The reason is mundane: the Mac went to sleep. You closed the lid, or you just didn’t touch it. Either way the running process was suspended. The agent’s request died mid-flight and a long job has to start over.

This isn’t unique to agents. Builds, large downloads, video exports, backups, database migrations, model training — anything you kick off and walk away from shares the same trapdoor.

Default (sleep enabled) Lid closed / left idle Asleep — the job stays suspended Running Keep awake ON Running Keeps going with the lid shut, and it's done when you return
Sleep triggers on time since you last touched the machine — not on whether the machine is busy.

macOS doesn’t distinguish “unused” from “not working”

Mac sleep is driven by whether a human is interacting. No keystrokes, no trackpad, lid closed.

What the Mac is actually doing is largely not part of the decision. The CPU can be pinned and the network saturated; if the idle conditions are met, it sleeps anyway.

As power management, that’s a defensible default. Historically, an untouched Mac was an idle Mac.

But the agent era breaks that assumption. Nobody is touching it and it’s working hard. In fact “busiest precisely when nobody is touching it” is becoming the normal case.

What actually goes wrong with agent workloads

The nasty part isn’t that a long task gets interrupted. It’s that the interruption is easy to miss.

  • You return expecting completion and find it stalled partway
  • The API connection dropped without producing a clear error
  • Half-written files are left behind and it’s unclear how far it got
  • A long job has to be rerun from the beginning
  • On metered APIs, you already paid for the part that got thrown away

That last one stings. With token-billed models, the interrupted work still bills, and rerunning means paying twice.

There’s a morale cost too. The real loss might be thirty minutes, but “I have to wait another thirty minutes” lands heavier than the clock says.

One toggle

MacSide AI’s keep awake solves this in a click. While it’s on, the MacBook won’t sleep — including with the lid closed.

The workflow is boring, which is the point:

  1. Flip it on before starting the long job
  2. Close the lid, walk away, or work on something else
  3. Flip it back off when the job finishes

Same goal as running caffeinate in a terminal, except there’s no command to remember and no terminal window you have to avoid closing. It’s a switch in a panel.

And it isn’t meant to be left on permanently. Turn it on when you need it, turn it off after. Keeping that toggle cheap is what makes it get used — anything heavier and you’d skip it “just this once,” which is exactly when it bites.

Clamshell mode benefits the same way

Outside agent workloads, the biggest use is clamshell mode: MacBook closed, driving an external display and keyboard. Tidy desk, bigger screen, very common setup.

macOS behavior here depends on power and display connection state, and when the conditions aren’t all satisfied, closing the lid sleeps the machine. That’s the origin of the reseat-the-cable, check-the-charger ritual.

With keep awake on, you stop negotiating with those conditions.

Where it earns its keep

  • Long AI agent tasks — walk away and the work continues
  • Big builds and compiles — leave a 40-minute build alone
  • Video export and rendering — stay awake to completion
  • Large downloads or uploads — run them overnight on a slow line
  • Backups and first-run sync — anything measured in hours
  • Clamshell with an external display — keep working with the lid shut
  • Waiting in a video call — the screen doesn’t drop during a long silent stretch
  • Extended screen sharing — no awkward sleep mid-explanation

Frequently asked questions

Should I leave it on all the time?

Better to enable it only when needed. Sleep exists for power saving and battery health, so there’s no benefit to disabling it permanently. Flip it on for a long job, off when the job’s done.

Does it really keep running with the lid closed?

Yes — while it’s on, the Mac doesn’t enter sleep, closed lid included.

What about heat?

A closed lid is a worse thermal situation. For long heavy jobs, put the machine somewhere with airflow — and if you can, prefer leaving the lid open with the display off rather than closed.

How is this different from caffeinate?

Functionally close. The difference is friction: nothing to memorize, no terminal window you must leave open. It’s a toggle.

Can it replace Amphetamine?

For simply preventing sleep, yes. If you need fine-grained triggers — auto-enable when a specific app launches, conditional rules — Amphetamine goes deeper. Keep awake is one of many features inside MacSide AI.

It was stopped the whole time you were waiting

The more work you hand to agents, the more hours your Mac spends working without a human present. Issue an instruction, walk away, come back later. That pattern is only going to get more common.

macOS sleep still reads “nobody is touching this” as “nobody is using this,” and stops the machine mid-job when the conditions line up.

Keep awake is a small feature that temporarily overrides that reading. Not impressive. But anyone who has waited thirty minutes for something that actually died after five knows exactly how practical it is.

One switch before you start. That’s all it takes for waiting to actually mean waiting.