In How I Run Three AI Models in Parallel Without Losing My Mind, Katie Parrott at Every brilliantly describes a problem I have too: working with AI makes me multitask like a pinball.
Katie includes practical tips in her piece on how to deal with the problem, like limiting your active tasks to three and dropping one line notes (“breadcrumbs”) about what to do next for the current task before switching to another.
Still, I’m convinced this multitasking craziness is a losing game.
The Problem with Multitasking
Over the past two decades, I’ve read pretty much every article, book, and study1 there is about multitasking and how it undermines your productivity2.
There’s a huge cost in switching between tasks paid in fatigue, stress, and time. The most important facts:
3% of people thrive when multitasking, everyone else’s performance suffers.
The more you multitask, the more you believe you’re being more productive, but you’re not.
The side-effects of multitasking apply to deep work but also smaller tasks. Even glancing at a phone notification is multitasking and costs you3.
Switching sometimes helps when you’re truly stuck, but that point usually comes after hours, not seconds or minutes.
In It Doesn’t Have to Be Crazy at Work, Jason Fried and David Heinemeier Hansson beautifully express this idea:
There are lots of ways to slice 60 minutes.
1 × 60 = 60
2 × 30 = 60
4 × 15 = 60
25 + 10 + 5 + 15 + 5 = 60
All of the above equal 60, but they’re different kinds of hours entirely. The number might be the same, but the quality isn’t.
The quality hour we’re after is 1 × 60.
Now, why does AI encourage this behavior?
As you become more advanced in your AI usage, you’re likely going to give multiple tasks to multiple AI models and agents, because they take time to complete their assignments.
Instead of waiting around for one to complete, you get it started, give another task to a second AI, and even a third. Then you return to the first to check its work, then the second, and so on.
Katie compares this to the work of a manager (and the video game Overcooked, in which you “play line cooks in a chaotic kitchen, sprinting between stations while orders pile up and the clock ticks down.”)
Seconds vs. Minutes vs. Hours
I can’t speak to the video game, but the manager comparison is accurate, except for one critical difference: managers don’t have to chunk up their time in seconds or minutes (though many do). They wait for hours or days to get back results from their reports.
With AI, your feedback loop is usually minutes.
Herein lies the real problem: minutes are too long to wait around and do nothing, but too short to avoid the averse effects of hyper-multitasking.
And so you arrive at the situation Katie describes: on a good day, you feel productive4, but probably still pretty tired from all the switching. On most others, everything is messy and you’re hungover from task-bingeing.
Why this multitasking problem might disappear soon
My hope is that this multitasking problem will go away by itself, and perhaps you’ll never experience it, for two reasons:
Most people aren’t yet running multiple AI agents or Claude Code sessions in parallel.
With faster, more capable AI, the “minutes” middle disappears: single tasks finish in seconds, while agents handle whole projects and run for hours, so you can work on a manager-style cadence.
We’ve then reached the same state managers of human reports now experience: you can collaborate on a single task together with your AI without the wait times, or assign them a project and get back the results hours later.
This cadence gets us back to a more natural rhythm: one task at a time, or at least 10‑minute‑plus timeboxes that prevent the worst effects of multitasking.
“You can do so much in ten minutes’ time. Ten minutes, once gone, are gone for good. Divide your life into ten-minute units and sacrifice as few of them as possible in meaningless activity.”
— Ingvar Kamprad, Swedish business magnate, founder of IKEA
Still convinced you’re a great multitasking?
You might belong to those 3% of super taskers, but that chance is, well, three out of 100. Here’s my final attempt to convince you otherwise:
Watch this brilliant explanation by Henrik Kniberg on the problem and cost of context switching.
Do this simple exercise I recorded in my previous life. It takes a few minutes and you’ll experience exactly what happens when you multitask (and find out if you’re a super tasker or not!).
Until AI catches up to either instant responses or true autonomous project management, we’re stuck in this maddening middle ground.
The best we can do is recognize the trap and try Katie’s breadcrumb approach, or simply accept that some days we’ll feel like line cooks in that chaotic Overcooked kitchen. At least we’re not alone in the madness.
Quick Links
🎧️ Just finished listening to the Google - The AI Company episode of Acquired. Amazing, amazing, amazing. It’s essentially a history of how we got to the current AI revolution, with unbelievable anecdotes and details.
🤖 I created a (blog) style guide generator with Claude Code, then turned it into a plugin a few days later so you can easily install it into your own Claude Code environment.
I was first introduced to the idea that multitasking doesn’t work through Getting Things Done. BrainChains is packed with research about this topic. Meet the Super Taskers is also a classic and fascinating read.
As long time subscribers know, I founded a productivity hardware + software startup in a previous life.
So it’s not just creative folks who suffer from this. Even managers do much better when they work in 10 or 15 or 20 min. intervals vs. switching every few minutes or seconds.
But that might still be an illusion, as mentioned earlier in the article: “The more you multitask, the more you believe you’re being more productive, but you’re not.”