A couple months ago I wrote about being more productive than ever while also being exhausted. Managing a fleet of agents across 10-15 projects a day. Learning faster than ever. Shipping constantly. Tired in a way I’d never been tired before.

I know. We’re all vibe coding. We’re all more productive than ever. This is your friendly reminder that we are not capable of multitasking and context switching is a recipe for burnout.

Context switching is the worst way to get through a day.

The CI/CD day

The most brutal days I’ve had as an engineer are almost always building CI/CD pipelines. Make a change, push it, watch the pipeline, see what happens. You get slightly further each time, but something always breaks. You look up and the day is gone. You start rigging ways to test the pieces that are failing without rerunning the whole pipeline. Hacking shell scripts together. Commenting out stages. Whatever it takes.

It’s a slog of a day. But when it’s finally done, you have ONE thing done. It’s solid. You understand the system now. You earned it.

The temptation with Claude is to work on that slog alongside five other things. And now you’re slogging and you have five half-finished projects at the end of the day. How do you feel tomorrow? You wake up, inventory half-baked work from yesterday, try to rebuild context across all of it, and you’re already behind before you’ve started.

Surely this can’t be my new life as an engineer.

Vaporwave and staring at the wall

I started thinking what if I went back to one focused task. Since we’re all riding the nostalgia wave anyway, why not try the old methods of development.

One. Focused. Task.

If I’m bored, be bored. I find a vaporwave album. Raise my desk to standing. Grab a couple things to fidget with—a ball, a plushie to toss to myself. Leave the window open. If I’m waiting on a build, I’m waiting. I stare at the wall. I bounce a ball. I watch the screen tick by.

It doesn’t feel more productive. My brain wants me to open another terminal and kick something else off.

Don’t.

Actually finishing things

I’ve been doing this for a couple of weeks. One day I looked up and realized I was less tired.

I was completing tasks. Not touching tasks. Not making progress on tasks. Completing them. Done. Merged. Shipped.

My end-of-day summary went from “touched these 12 things” to “finished these 2 things.” Two finished things beats twelve touched things every single time.

The mental overhead of tracking unfinished work was worse than chrome tabs. Instead of an article I would maybe someday read, it was 12 tasks all half-finished in weird states. Each one needing me to rebuild context before I could make progress. It was leaving me exhausted before I even started working.

I still get the productivity boost. The AI is still doing its thing. It’s just concentrated instead of scattered. Better context in, better code out. Less drained at the end of the day.

Let the AI slog for you

Bonus points if you figure out how to give the AI the ability to test whether what it did actually worked.

For the CI/CD slog days, the real unlock is closing the feedback loop. Instead of push, switch to something else, come back later, forget what you were doing—let the agent watch its own pipeline:

# Push and wait for checks to complete
gh pr checks <pr-number> --watch

# Or watch a workflow run
gh run watch

The agent pushes, waits, reads the failure, fixes it, pushes again. You’re still focused on one thing, but now the iteration loop is tighter. The slog day becomes a slog morning. And you were only ever working on one thing, so when it’s done, it’s actually done.

The wall is doing something

The gaps between tasks aren’t wasted time. Your brain is processing. AI made every gap fillable. Filling them all is how you burn out.

If code is now cheap, then so are prototypes. Perhaps the mental load of juggling unfinished work is more expensive than the code we generated when it wasn’t our primary task.

Fewer things. Done completely. Room to breathe. Walk away at the end of the day.