The Loss of Flow

How AI Killed the Best Hours You Ever Had at Work

The Craftsman

Picture a cabinetmaker in the early 1800s. He's been working since dawn – hand on walnut, chisel meeting grain, perfecting the fit of a dovetail joint for a drawer in an oversized apothecary chest. He couldn't tell you what time it is. It wouldn't even occur to him to wonder. Hours feel like minutes. His sole focus is on his creation taking form.

He's not happy, exactly. He's something rarer than happy.

He's absorbed.

He's experiencing flow.

Now picture the same guy, except this time, we've mercilessly time-traveled him 100 years later. Here, we find him on a sprawling furniture assembly line. Here, he also works with wood. Technically.

His job is to monitor a lathe, check a measurement, pull a lever, and wait for the next piece to come down the line. He does this every ninety seconds, like clockwork. Now, he can tell you exactly what time it is, because his work isn't a continuous act of creation. It's a loop. Stimulus – response – stimulus – response ad infinitum. Morning whistle to evening whistle.

You may recognize this rhythm.

---

If you do any kind of knowledge work in 2026, your day has probably developed into something like this: You write a prompt. You wait. A result appears. You scan it – small spark of satisfaction, or flicker of disappointment – and then you write another prompt. While it's running, perhaps you check Slack or look at the notifications on your phone. Then back to Claude or ChatGPT or Gemini – sometimes all three. Maybe you do this twenty times before lunch. Or thirty, or fifty? Each cycle burns a few minutes here and there. Each one carries a tiny dopamine hit of anticipation and reward.

By any external measure, you are enormously productive compared to a year ago.

But something has changed about the texture of your work, and it's worth being precise about just what that is.

The Slot Machine That Feels Like Work

That cycle – prompt, wait, evaluate, repeat – is what behavioral psychologists call a "variable ratio reinforcement schedule." It's the same mechanic that makes slot machines so compulsive, and is often referred to as a Skinner Box, dredging up imagery of a rat mindlessly pressing a lever for the chance of food.

(Even though it's over 20 years old, one of my favorite game design articles, Behavioral Game Design, is worth reading. Now, its relevance has broadened beyond games to any app in the attention economy, and it will give you some insight into how apps manipulate your motivations.)

Unlike a slot machine, though, the prompt-check loop doesn't feel like a vice, does it?

It feels like productivity.

You don't have a guilty conscience because you're getting things done! You're producing new articles! You're getting your contract reviews done 10x faster! You're writing tools that will compound your productivity and make you even more effective! You're shipping.

The reinforcement pattern is wrapped in accomplishment, which makes it almost impossible to notice what it's quietly replacing.

It's all insidiously something I think of as productive distraction.

This productive distraction leads to the loss of flow.

And it can also lead to "AI Brain Fry" if you sink too deeply into the quagmire. But today, I'm mainly disturbed by the loss of flow, so that's what I'm focusing on.

Flow

Most of you are probably already aware of "flow," but I want to dissect it a bit. It's important.

In the 1970s, a psychologist named Mihaly Csikszentmihalyi interviewed a motley crew of artists, athletes, surgeons, and chess players about the moments when their work "felt best."

His findings were remarkably consistent. Across all these disciplines, people described the same thing: a state of total absorption where time distorts, self-consciousness fades away (a hint of losing the ego and becoming one with the universe), and the work seemed to move through them rather than being deliberately muscled into existence. He called it "flow."

Physiologically, flow is intense exertion. But it doesn't feel like exertion. It feels like alignment – what you can do matches so precisely to the difficulty curve of what you're doing that it leads to a flowing state of blissful accomplishment.

It's not uncommon to find that suddenly hours have passed, you've finished your task, and you feel a great sense of satisfaction. Almost like coming out of a trance.

The catch is that flow has a steep entry price. It takes 15-20 minutes of unbroken focus on a single task before the state begins to take hold. Interruptions reset the clock. Notifications from your phone or work computer are major culprits. I've personally taken to silencing all Slack notifications at work unless I'm specifically mentioned, and even then, it can be too disruptive at times. Not to mention if you become a slave to notifications, your brain starts anticipating them, and develops a notification loop of its own where it thinks every few minutes: "Maybe I should check for notifications now? Glance at my phone? Switch over to Slack?"

This context-switching leaves what the researcher Sophie Leroy calls "attention residue" – cognitive traces of the prior task that degrade your presence in the next one after switching contexts. The more switches, the more attention residue.

Let that phrase sit with you a moment so you might take it with you today.

Attention residue.

The prompt-check loop, with its two-to-four-minute cycles can create massive attention residue. And how many simultaneous agents do you have running to context switch among, might I ask?

It's almost perfectly engineered to prevent flow from ever starting.

And if flow never begins, what do we lose?

From Chef to Taster

The strange thing is that AI-assisted work doesn't feel mindless. It still requires skill. Skills of judgement, of taste, of editorial instinct, of direction. So why doesn't it feel the way deep work used to feel?

Well, notice something about all of those skills?

They're all supervisory skills. They're the skills of a manager reviewing a direct report's work.

What are they not?

They are not a maker building something with their own hands.

Flow suffocates in this environment.

Think about the difference between a taste-tester and a chef. The person sampling tastes of a dish and saying "needs more tomatoes" or "too much salt" is doing something useful. It takes a good palate, real judgment, and sometimes decades of experience. But they're not the one standing over the heat – building the dish from nothing into something.

The chef is.

And if you're thinking, "But I AM the chef! The AIs are my line cooks! I'm directing the dish, and the vision is mine." Okay, fair. But consider this. When a chef gets successful enough that they stop actually cooking and start managing the kitchen full-time — tasting, directing, sending plates back — even though the food is still theirs and the menu is their vision, many of them describe a real sense of loss. Because the work that made them love the profession wasn't the directing. It was the cooking. The heat. The focus. The absorption. The flow of creation.

You can be the chef and still lose the best part of being one.

The renowned chef Anthony Bourdain was famously devoted to the physical act of cooking and resisted the pull toward pure management. In his words: "I'll be right here. Until they drag me off the line. I'm not going anywhere."

AI has turned a lot of knowledge workers into taste-testers of their own output. It sounds like a great deal until you realize that most of them got into the profession because they wanted to cook.

The Plate Spinner

If the prompt-check loop is the new assembly line, then the person running a dozen AI agents simultaneously is the floor manager overseeing multiple lines at once.

And this has become – you've seen the LinkedIn posts – aspirational. People brag about it. You've got one agent drafting a report, another analyzing your data, a third doing research, others triggered every hour by cron jobs to generate summaries of random things you won't read anyway, your dashboard looking like air traffic control on a day when the government is well-funded. You're an ER physician doing triage: Which agent needs redirection? Which result just came in? Which task stalled?

"Hmm... maybe I'll vibe code a tool to coordinate all that," you think.

I'd be kidding myself if I didn't admit to feeling this tug myself. I love AI. I've used it for music, writing, art, code, business plans, spreadsheet formulas, data aggregation, exploring philosophies of life and existence, and who knows what else? And it's this reflection on what it's doing to me that has brought me to writing this piece. I think this piece may be more for myself than it is for you. But maybe if we all hear this together, we'll escape with some of our creative humanity intact.

The Colonization of Sleep

There was one boundary that productivity culture could never cross: Sleep.

Sleep was an eight-hour ceasefire.

AI agents dissolved the boundary. Count yourself lucky if they haven't crept into your closets at night to spring out at you when the lights go off.

They attack from the darkness even if you don't believe in monsters. If an agent can run overnight – analyzing, generating, preparing tomorrow's inputs – then every hour it isn't running represents lost output. That's scary.

Sleep becomes a period of waste.

When you roll over at 3 am, see a notification that your agent crapped out, and feel that creeping need to restart it lest you be haunted by the thought of unused capacity – well, sleep is no longer a period of rest.

It reminds me of my first job in the game industry as a gameplay scripter on The Sims, before Maxis and Electronic Arts had sane work hours. I crunched from breakfast through dinner and beyond into the silent night of a half-asleep office as I stared at our proprietary scripting interface ("EDITH") so much that it haunted my dreams. It was as if my brain felt like it had to be productive at all hours. That was the culture. And it was not restful sleep as abstract images of nodes and lines danced like whatever the opposite of sugarplums are through my head.

Although my early days on The Sims were an extreme, there's still a difference.

Worrying about your overnight agents is an anxiety born of optimization.

Remember the wise adage: Life is not about the destination, but the journey. If you don't enjoy the work – if it's just seeking a superficial hit of dopamine, or worse, being struck with loss aversion, then you are making happiness harder to find.

The Trade

Let me be honest about this, because the article falls apart if I'm not: AI-assisted work is more productive by almost any conventional measure. Faster, broader, more polished, more scalable. The person who has mastered orchestrating twelve agents will outproduce the solo worker in nearly every quantifiable dimension. The overnight agents will generate a real advantage by morning.

Flow can't compete on efficiency. It loses that fight.

Instead, this is an argument about the inner life of the person doing the work.

The ancient Greeks had two words for happiness: hedonia – the pleasure of the moment – and eudaimonia – the deeper fulfillment that comes from exercising your abilities in pursuit of something meaningful. The prompt-check loop is hedonic. Small, frequent rewards that feel good in the instant and leave little trace. Flow is eudaimonic. Intense, consuming, hard-won – and the source of the deepest contentment people report in their working lives.

We are, collectively and without quite deciding to, trading the second for the first.

And this is the struggle: AI tools made the shallow path so rewarding on a minute-by-minute basis that the deep path of flow got harder and harder to justify. If your peers and competitors are working at three times your speed with AI, choosing to do the work yourself is a choice of purity, but it's reckless.

It's a structural shift, and like the assembly line before it, it is very nearly impossible to opt out of once it becomes the standard.

It's also not lost on me that as I come to the close of this piece, I'm emerging from a multi-hour state of flow. And it feels really, really good. The feeling is an old friend I've not seen since late last year, before the din of the holidays. Yet I distractedly look over his shoulder... and I feel the slightest nagging thought about how many other things I could have accomplished in this time, had I had agents running or had I just fired off AI to write this piece for me instead.

---

Now, a final vignette.

You sit at a desk, monitors aglow, morning light streaming through your window. On one screen, an agent generates a first draft for you. On another, results from an overnight run wait for your review and approval. A third shows a chat interface, cursor blinking, ready for your next prompt.

You take a sip of your favorite morning drink. Scan the output. Start typing a new instruction.

You're productive. You're in command. You will finish more today than you once completed in a week, or a month.

But you will not, at any point in the hours ahead, look up and realize the afternoon has just vanished from absolute absorption in your craft. You will not lose yourself in your work.

You'll be present for every moment of it.

And that, quietly, is the problem.

At some point, between prompts, in one of those few-minute windows while you wait for a result, you'll feel a small, hollow restlessness — not quite boredom, not quite dissatisfaction, just a vague sense that something is missing. You'll check your phone. You'll tab over to Slack. The result will come back. And the day ticks forward.

Somewhere, the cabinetmaker works on.