If AI coding leaves you drained, foggy and bored at the same time, you're not just being whiny. It's a real, measurable thing, and it comes from how the human brain handles the kind of work AI coding has turned into. The good news is that you can get some of your agency, flow and joy back, and it doesn't take more willpower.
The short answer: AI coding feels bad because your willingness to put in effort drops before your ability does, running several agents overloads your working memory, every switch between unfinished tasks leaves "attention residue," agent work breaks the conditions for flow, and supervising AI instead of building slowly erodes your sense of agency. What helps is a trigger-based mindfulness framework called AIM (Awareness In Motion), created by Lewis Bates: set your intention before you start, do a quick check-in every time you switch tasks, and watch both the output and yourself in between. This guide covers the research behind why it feels so bad and exactly how to use the framework.
The short version
| Why it feels bad | What's going on | What to do about it |
|---|---|---|
| "AI brain fry" | Mental fatigue from overseeing AI beyond your cognitive capacity | Notice fatigue earlier with check-ins |
| Effort drops before ability | Tired brains avoid hard thinking while performance still looks fine | Check how you feel, not just how you're performing |
| Working memory overload | About four chunks; every agent has its own context | Slow down by ~10% and keep fewer things in flight |
| Attention residue | Part of your attention stays on the unfinished task you left | Check in at every switch; leave a "ready-to-resume" note |
| Flow disappears | Slow feedback; each action too easy, total load too much | Set a clear goal before you start |
| "What am I for?" | The job shifts from building to supervising | Decide who the work is for and what standard you hold |
How AI coding took the fun out of programming
I hate AI coding. And honestly, I'm really sad about what it's done to the way we do our jobs.
I fell into this field by accident. I was doing a master's in public policy, and in 2018 I took my first programming class, R for statistical analysis, and I just loved it. I loved getting stuck on a puzzle with no way out except Stack Overflow for hours, sometimes days, and then that dopamine hit when it finally worked.
As I got better, I fell in love with the flow state. I'd put on some techno or instrumental metal, blast through code for hours, and end the day feeling energized. I never really cared about the end result. I just like building stuff.
When AI first showed up, I actually really liked it. It replaced the hours I was spending digging through Stack Overflow, so the dopamine hits came faster and I was still in that flow.
But over the last year or so, the models got really good. I wasn't programming anymore. I was managing a bunch of agents in parallel. Sure, I was producing more code than ever, and pretty good code too. But I couldn't listen to music anymore. I'd sit there in silence with a bunch of terminals open, constantly context switching, with these weird gaps where I'm waiting on an agent and don't know if I should switch to another one or just sit there. I was bored and overwhelmed at the same time, and the flow state was just gone.
It cost me more than I realized at the time. For most of the last two years since I've been making content, I was also working full time at Twitch. When I loved my job, I still had the energy to work on content on the weekends. But once my days were all AI coding, I'd end the day and the week completely empty and braindead. I think it's honestly one of the reasons I quit my job at Twitch in September. I'm really grateful to have something else I'm passionate about to turn to.
Not everyone feels this way
My dad has been programming since the eighties, and for him AI coding is awesome because it finally got rid of all the annoying stuff. A lot of my coworkers loved it too.
But for me, the problem solving was the fun part, and I don't think I've really let myself grieve that. I've only been at this for eight years. I can't imagine spending thirty years perfecting your craft and then having it basically taken away. If AI writes the code, what am I for?
People say, just code by hand as a hobby. But now I'll always know I could ask AI and have the answer in a minute. The grind used to be necessary, and I don't think it's going to feel the same when it's optional. And with code, handmade doesn't really count for anything. If an expert woodworker builds you a table, the work that went into it is part of why it's valuable. Nobody cares whether the app they're using was artisanally hand-coded.
Why being sad isn't the end of the story
There are so many videos and posts right now about how AI ruined programming, and I agree with a lot of them. But they usually end there, with being sad. I get it, but I think stopping there takes away our agency. AI coding is here to stay, and most people can't just quit and become an educator like I did. We need a way through this that isn't just giving up.
One of my coaching clients, Lewis Bates, was going through the same thing. Lewis is a staff engineer in medical AI. After a few days of really intense AI-assisted work, he noticed he was foggy, flat, tired, and thinking more slowly. And he asked himself a pretty relatable question: is the work I find productive and meaningful actually harming me?
So he went digging through the research and built a system for working with AI that doesn't fry your brain, called AIM (Awareness In Motion). He's kindly let me share it here. First I'll walk through why AI coding feels so bad, because the science is really interesting, and then what we can actually do about it.
Why does AI coding feel so bad?
So, is this a real thing or are we all just being kind of whiny? I'm pretty sure it's real.
"AI brain fry" is real
Earlier this year, researchers from BCG surveyed 1,488 full-time US workers and published the results in Harvard Business Review. Among the people using AI at work, 14% said they'd experienced what the researchers call "AI brain fry," which they define as mental fatigue from using or overseeing AI tools beyond your cognitive capacity. People described a buzzing feeling, mental fog, trouble focusing, slower decisions, and headaches.
- The people doing the most AI oversight, which is basically what running a bunch of coding agents is, reported 12% more mental fatigue and 19% more information overload.
- Self-reported productivity went up for people using three or fewer AI tools, then dropped off once they were juggling four or more.
- People with brain fry were more likely to be planning to quit: 34%, compared to 25% of everyone else.
Your willingness to put in effort drops before your ability does
So what's actually going on in our heads?
A team at Johns Hopkins looked at this last year (Journal of Neuroscience). They put people in an fMRI scanner and had them do a hard working memory task over and over. In between rounds, they offered a choice: do a harder version for more money, or an easier version for less.
As people got more tired, they turned down the harder options more and more, even when the money went up. Their performance didn't drop at all, though. They could still do the hard task just fine. They just didn't want to anymore.
The researchers saw two brain regions ramp up as fatigue built: the dorsolateral prefrontal cortex, which handles working memory, and the insula, which is involved in how we feel fatigue and weigh whether effort is worth it. When people were tired, those two regions were more connected during their choices, and the insula got more sensitive to how much effort something would cost. The researchers' interpretation is that the effort signal from your prefrontal cortex feeds into that "is this worth it" calculation.
This is the part Lewis zeroed in on, and I think it explains a lot. Our willingness to put in effort drops before our ability does. So brain fry shows up first as "ugh, I really don't want to read this huge PR," and only later as actual mistakes. And when your job is reviewing AI output, not wanting to read carefully is exactly the problem.
Running several agents overloads your working memory
Then there's all the switching between terminals. Working memory, the stuff you can actively hold in your head at once, is small. The psychologist Nelson Cowan estimated it at about four chunks of information. Every agent you're running has its own task, its own context, and its own half-finished state, so with three or four of them going, you're past that limit all the time.
Every switch leaves attention residue
And every switch costs something. Sophie Leroy, a business school researcher, found that when you move from an unfinished task to a new one, part of your attention stays stuck on the first one. She calls it attention residue, and it makes you worse at whatever you switched to. With agents, nothing is ever really finished. We're always leaving one halfway through to go check on another.
Why the flow state disappears
That's also a big part of why the flow state disappears. Mihaly Csikszentmihalyi, the psychologist who coined the term, found in Flow: The Psychology of Optimal Experience that flow needs a few things, including fast feedback and a challenge that matches your skill. Agent work messes with both:
- The feedback is slow. It comes after you've been sitting there waiting.
- The challenge is mismatched. Each individual action, like approving a change or re-prompting, is too easy, while the total load is too much.
That's how you end up bored and overwhelmed at the same time.
The "what am I for?" feeling
And then there's the "what am I for" feeling. Researchers at Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers who use generative AI at work every week. They found AI shifts the work from solving problems yourself to verifying, stitching together, and supervising what the AI produces. The more people trusted the AI, the less critical thinking they reported, while people who were more confident in their own skills did more of it.
Lewis describes this as a slow erosion of agency and competence. The job turns from building into supervising, and the part of your brain that had fun building things stops getting used.
Putting it together
- Our willingness to do hard thinking drops before we notice we're tired.
- Fluent output gets less scrutiny.
- Working memory is overloaded, and every switch leaves residue behind.
- The choice to think for ourselves turns into a habit of just... not.
That is just what happens to the human brain under these conditions, which means the solution can't just be "try harder."
What can you do about AI coding fatigue?
There actually is something that can help a lot here: mindfulness.
Bear with me. This isn't going to get woo-woo. Mindfulness just means paying attention to what's going on in your head and around you. And it fits this problem really well, because almost everything we just talked about is a noticing problem.
In the Johns Hopkins study, people's brains were registering fatigue and steering them away from effort while their performance still looked fine. Brain fry starts before we're aware of it, so the first skill we need is noticing it sooner.
The research here is pretty legit. A big review in Nature Reviews Neuroscience pulled together two decades of studies and found measurable effects on attention control, emotion regulation, and self-awareness. Attention control and self-awareness are exactly what all that switching and fatigue chip away at. And it's trainable, the same way other skills are, because our brains change based on what we repeatedly pay attention to.
It runs on triggers, not willpower
The thing I like most about Lewis's approach is that it runs on triggers instead of willpower. Remember the Johns Hopkins study: when you're tired, your brain gets less willing to spend effort, and remembering to check in on yourself takes effort. So every piece of this is linked to something you already do throughout the day, and you don't have to remember to be mindful in the moment.
| When | What you do | What it targets |
|---|---|---|
| Before you start | Decide who benefits and what standard you care about | "What am I for?", lack of a clear goal |
| Mid-task and before you ship | Check the work still serves that intention: would I put my name on this? | Eroding judgment |
| Every time you switch tasks | Three-breath check-in on body, head and mood; leave a resume note | Hidden fatigue, attention residue |
| In between | Watch the output and yourself; slow down ~10% when the pace takes over | Working memory overload, autopilot reviewing |
Step 1: Set your intention before you start
Before you get started with work, take a minute and answer two questions:
- Who actually benefits from this work? Maybe it's a customer who needs this feature, or a teammate who's going to maintain this code.
- What standard do you care about holding?
For me it's simple. I care about the quality of my work because I feel like it represents me, and I don't want to output crap. Even a value that basic is enough to keep me invested in the work itself.
Then come back to it while you work. Once in the middle, to check whether what you're building still serves your intentions and still meets your bar for quality. And once right before you deliver the code, with the question: would I put my name on this?
This is aimed at that "what am I for" feeling. Remember the Microsoft and Carnegie Mellon study: the people who were more confident in their own judgment did more critical thinking. AI can write the code, but deciding who it's for and what good enough looks like is still our job, and making that call deliberately keeps our judgment in the loop. It also gives you a clear goal, which is one of the things flow needs in the first place.
Step 2: Check in every time you switch tasks
Knowing why you're doing something doesn't stop the fatigue. You can have a clear goal and still get pulled along by the pace of the agents until your brain is mush. So the next question is how to catch it early, while it's still at the "I don't feel like it" stage, before it turns into mistakes.
The answer is a really short check-in every time you switch tasks: when you're opening an AI tool, switching to a different agent, or coming back after waiting on one. Those moments are exactly where attention residue builds, and you already make them constantly, so they're the perfect trigger.
The reason you check how you feel is that Johns Hopkins finding again: fatigue showed up in how people felt and what they were willing to do long before it showed up in their performance. So feeling tense, scattered, or irritable is your early warning, and you'll notice it way before you start actually performing worse.
Before you switch, take a breath and ask yourself three quick questions:
- Body: Am I tense? Is my jaw clenched, or are my shoulders up by my ears?
- Head: Is my head clear or scattered?
- Mood: What's my mood right now: curious, impatient, annoyed?
You're not really trying to fix anything. You're just checking your own state before you dive into the next thing. Once you've done it a few times, it takes about three breaths and becomes pretty automatic.
One more thing that helps here, from Sophie Leroy's later research: before you leave a task, jot down a quick note on where you are and what you'll do next. In her experiments, that kind of "ready-to-resume" plan reduced attention residue and helped people do better on whatever they switched to. With agents, that can be some notes in a scratch file before you jump to the next terminal.
Step 3: Watch the output and yourself in between
A lot of the drain happens in the long stretches between switching tasks, like when you're deep in one agent's output and the pace just carries you along. So the last piece is something you keep doing the whole time you're working. You keep an eye on two things:
- The output. Is it right? Does it meet the spec? Am I actually reading this diff, or just scrolling past it?
- You. Am I still present, or has the agents' pace taken over?
When you notice the pace has taken over, take a breath and slow down by about ten percent. Slowing down usually means fewer things in flight at once, which keeps you closer to what your working memory can actually handle at any given time. Remember, the BCG survey found productivity dropped off once people were juggling four or more AI tools. And if slowing down isn't enough, go back to the check-in before you keep going.
How it all fits together
You decide what you care about before you start and check it again before you send a PR. You do a quick check-in at every switch. And in between, you watch the output and yourself, and that tells you when you need another check-in.
Can AI coding feel good again?
This isn't going to make AI coding feel exactly like my good old days of blasting Russian Circles and grinding through Stack Overflow. That part is gone, and I think it's okay to be sad about it. But how we feel while we do this work isn't totally out of our hands.
AI coding is here to stay, and a lot of us want to stay in this field we love, so I'd rather find a way to enjoy it than give up.
If part of what's weighing on you is whether AI is going to take your job entirely, I went through what the data actually says in another video.
And if you want help building your own production-ready project, I run a small, application-only build cohort where I personally work with a handful of people over eight weeks to take an AI engineering project all the way through deployment.
Frequently asked questions
What is AI brain fry?
AI brain fry is a term from researchers at BCG, published in Harvard Business Review in 2026. They define it as mental fatigue from using or overseeing AI tools beyond your cognitive capacity. People described a buzzing feeling, mental fog, trouble focusing, slower decisions and headaches. Among 1,488 surveyed US workers, 14% of those using AI at work said they'd experienced it.
Why does coding with AI agents feel so exhausting?
Several things stack up. Your willingness to do hard thinking drops before your ability does, so you stop reading carefully before you notice you're tired. Running several agents at once overloads working memory, which holds about four chunks of information. Every switch between unfinished tasks leaves attention residue behind, and supervising AI output instead of solving problems yourself slowly erodes your sense of agency.
Why is it so hard to get into a flow state when coding with AI?
Flow needs fast feedback and a challenge that matches your skill. Agent work breaks both: the feedback comes after you've been sitting there waiting, and each individual action, like approving a change or re-prompting, is too easy while the total load is too much. That's how you end up bored and overwhelmed at the same time.
Does running more AI agents in parallel make you more productive?
Not past a point. In the BCG survey, self-reported productivity went up for people using three or fewer AI tools, then dropped off once they were juggling four or more. People doing the most AI oversight also reported 12% more mental fatigue and 19% more information overload.
How can mindfulness help with AI coding fatigue?
Mindfulness just means paying attention to what's going on in your head and around you, and AI coding fatigue is mostly a noticing problem: it starts before you're aware of it. Research reviewed in Nature Reviews Neuroscience found measurable effects on attention control, emotion regulation and self-awareness, which are exactly what constant switching and fatigue chip away at. The framework in this guide ties short check-ins to things you already do, so it runs on triggers instead of willpower.
What is the AIM framework for AI-assisted work?
AIM (Awareness In Motion) is a framework created by Lewis Bates, a staff engineer in medical AI. It has three parts: set your intention before you start (who benefits and what standard you care about) and revisit it mid-task and before you ship; do a quick three-question check-in on your body, head and mood every time you switch tasks; and in between, keep watching both the AI's output and yourself, slowing down when the agents' pace takes over.
How do you reduce context-switching costs when working with multiple AI agents?
Before you leave a task, jot down a quick note on where you are and what you'll do next. In Sophie Leroy's experiments, that kind of 'ready-to-resume' plan reduced attention residue and helped people do better on whatever they switched to. With agents, that can be a few notes in a scratch file before you jump to the next terminal. Running fewer things in flight at once also keeps you closer to what your working memory can handle.





