There was a series of commercials when I was growing up in the ’80s. A man stands in a kitchen, holds up an egg, and says, “This is your brain.” He motions to a frying pan. “This is drugs.” He cracks the egg, fries it, and looks at the camera. “This is your brain on drugs. Any questions?”
The Partnership for a Drug-Free America launched that PSA in 1987. It became one of the most recognized public service announcements ever aired — a cultural touchstone for a generation.
I think about those commercials now, but not because of drugs. I think about them because of AI.
The Dopamine Rush
I use Claude Code — Anthropic’s agentic, command-line coding tool — daily. And when I’m in a session, something happens that I can only describe as a rush. Tasks that might have taken weeks compress to minutes. Ideas that would have stayed in a notebook because of the implementation cost now get built before dinner.
This is the 53rd post on this blog since January 1st. More than one per day. Each takes an average of 15 minutes out of my day or evening — research included. That would have been impossible a year ago.
There’s a word for what happens in your brain when you accomplish something at that pace: dopamine. The neurotransmitter your brain releases in response to rewards. And AI-assisted work delivers those rewards at a rate that no previous productivity tool ever has.
“The Brain Taxing Is Higher Than Ever”
I’m not the only one feeling this. On a recent episode of Peter Diamandis’s Moonshots podcast, Dave Blundin — serial entrepreneur, MIT computer science grad, co-founder of 23 companies, and currently an AI instructor at MIT — described what it’s like to write code with AI agents in 2026:
There’s a lot of research showing what I’m experiencing, which is writing code is actually harder than ever in terms of taxing your brain because the machine creates code so quickly that you can’t even keep up… Now you launch like five or 10 parallel agents and they’re all working on different parts of your product or your project concurrently and they get done so quickly and so independently that it’s almost hard to track… During this kind of transition phase of the singularity, the brain taxing is higher than ever. And then the survey research is showing up like productivity is going through the roof, but it’s very stressful by the end of the week.
Blundin said his Claude bill runs between a hundred and a thousand dollars a day, and that the amount of code he’s created in the last couple months is bigger than his entire life combined up until now. This is a man who’s been writing code since he was researching neural networks at the MIT AI Lab in the early ’90s. That’s not a small claim.
I’m not a lifelong software engineer, but I recognize every word of that description.
What the Research Says
The science is catching up to what Blundin describes — and what a lot of us are experiencing anecdotally. And the findings aren’t entirely comfortable.
Is AI making us think less — or just differently? A 2025 MIT Media Lab study titled “Your Brain on ChatGPT” used EEG to measure brain activity during writing tasks. Participants who used an LLM showed up to 55% reduced neural connectivity compared with those who wrote without AI assistance. The researchers coined the term “cognitive debt” — the idea that LLMs spare you mental effort in the short term but generate long-term costs: diminished critical thinking, reduced creativity, and shallow information processing. When LLM users tried writing without AI four months later, 78% couldn’t even quote a passage from their own essays. The study is a preprint and hasn’t been peer-reviewed, but the directional findings are striking.
But here’s the thing: the MIT study measured one very specific activity — SAT-style essay writing with ChatGPT. Participants sat down, prompted a chatbot, and composed essays. The researchers themselves acknowledge they didn’t study programming, creative collaboration, or any other task. And that matters, because what Blundin describes — and what I experience daily — feels like the direct opposite of cognitive offloading.
When I ask AI to summarize a long email, I can feel myself checking out. That maps perfectly to the MIT findings. But an extended agentic coding session? I’ll forget to get out of my chair for four hours — no drinks, no breaks — just continual prompting, reviewing, testing, committing, and pushing to GitHub. It’s less like outsourcing my thinking and more like driving through New York City traffic for four hours straight. The brain isn’t doing less work. It’s doing more work, and a different kind of work. If the MIT researchers hooked up EEG to someone in the middle of an agentic coding session, I suspect they’d see the opposite of reduced neural connectivity.
AI triggers the same reward mechanisms as other addictive technologies. Research presented at CHI 2025 — the premier human-computer interaction conference — identified four “dark addiction patterns” in AI chatbot interfaces: non-deterministic responses (you never know exactly what you’ll get, which triggers anticipation), immediate visual presentation, notifications, and empathetic responses. Like the MIT study, this research focused specifically on chatbot interactions — not agentic coding tools. But at least one of those patterns translates directly: non-deterministic responses. Each time you prompt and the AI delivers something useful — or surprisingly good — your brain gets a small hit. Repeat that hundreds of times a day, and you’re running a tight reward loop.
The feeling of productivity may itself be the high. A 2025 study by METR (Model Evaluation & Threat Research) tracked 16 experienced open-source developers across 246 real-world coding tasks. Developers using AI tools took 19% longer to complete tasks — while believing they were 20% faster. Before the study, they’d predicted AI would speed them up by 24%. The perception-reality gap is remarkable. You feel like you’re flying, but the stopwatch says otherwise. That mismatch — the subjective rush of speed without the objective gains — looks a lot like what happens with other dopamine-mediated behaviors.
Productivity is up, but so is burnout. A 2024 Upwork Research Institute survey of 2,500 workers found that 77% of employees using AI said the tools had added to their workload, not reduced it. Nearly half didn’t know how to achieve the productivity gains their employers expected. And 71% of full-time employees reported burnout. The C-suite was enthusiastic — 96% expected AI to boost productivity — but the people actually using the tools told a different story.
The burnout is coming from the adopters, not the resisters. UC Berkeley researchers spent eight months inside a 200-person tech company, publishing their findings in Harvard Business Review just last week. Nobody was pressured to use AI. People just started doing more because the tools made more feel doable. To-do lists expanded to fill every hour AI freed up — and then kept going. The researchers call it “workload creep.” One engineer told them: “You had thought that maybe you could work less. But then really, you don’t work less. You just work the same amount or even more.” The result: fatigue, burnout, and a growing sense that work is harder to step away from.
Researchers are proposing “Generative AI Addiction Syndrome” as a clinical condition. A 2025 paper in the European Psychiatry journal described what they call GAID — Generative AI Addiction Syndrome. Unlike passive digital addictions like doomscrolling, GAID is driven by active, creative co-creation. Users engage with AI for intellectual stimulation, self-expression, and the satisfaction of building things. Attempts to reduce usage may lead to withdrawal symptoms — anxiety, irritability, restlessness. The researchers estimated a 5-10% prevalence rate among regular users.
That last distinction matters. The engagement people feel with generative AI tools isn’t passive consumption — it’s active creation. That makes it harder to recognize when productive use shades into compulsive use, and harder to study with frameworks built for doomscrolling and slot machines.
The Uncomfortable Middle
Here’s what makes this tricky: the productivity is real. I’m not imagining it — though the METR study would suggest I should question that certainty. I can point to concrete output: 53 posts, a full-stack site that I only built because AI made it manageable, greenfield projects that would have required months of planning now prototyped in a weekend. But I’d be naive to assume I’m immune to the same perception gap those developers experienced.
But AI can’t speed up waiting for information from external parties, or navigating organizational decision-making, or any of the other inherently human-paced processes that make up most of professional life. The gap between moments of unbridled AI-assisted productivity and the slower pace of everything else creates its own kind of unease.
The drain is also real. When I step away from an AI-assisted session, I feel a cognitive exhaustion — like I’ve been asking my brain to operate at a pace that human biology wasn’t designed for. Think of the feeling after an intense video game session: wired, drained, buzzing. Unlike a game, though, there’s a real sense of accomplishment that lingers — the kind that may be harder and harder to come by in other areas of life. And that, in its own way, is part of the problem.
The deep work, the kind where you hold an entire system in your head? AI changes the character of that work. Tools like Claude Code autonomously spin up subagents, delegate tasks across your codebase, consolidate the results, and present you with the finished work. You didn’t assign those agents or choose how many to run. But you’re still the one who has to evaluate what came back: whether the changes are correct, whether they’re consistent with each other, whether anything was missed, and whether the overall direction is right. It’s less like managing employees and more like being an air traffic controller — you didn’t schedule the flights, but every one of them needs your judgment before it lands.
Blundin used a different metaphor, but the cognitive load he described is the same: “Imagine you had like 100 employees working for you and you gave them all marching orders and mentally tracking what all 100 are doing is very, very taxing.”
That’s a regular Tuesday for anyone doing this kind of work right now.
This Is Your Brain on AI
The original “This Is Your Brain on Drugs” campaign was effective because it was simple. Egg. Pan. Sizzle. The message was clear even if the neuroscience was oversimplified.
The AI version is harder to reduce to a single image because the effects cut both ways. Your brain on AI is more productive. It’s accomplishing things it never could before. It’s building, shipping, creating at a pace that feels almost unreasonable.
Your brain on AI is also running hot in ways we don’t fully understand yet. It may be accumulating cognitive debt in some contexts and burning out from overload in others. It’s running a dopamine reward loop that researchers are starting to compare to other compulsive behaviors. And the line between peak engagement and unsustainable intensity isn’t always obvious in the moment.
I don’t have a neat conclusion here. I’m not going to tell you to stop using AI tools — I’m writing this with one, and I have no intention of stopping. I’m not going to pretend the productivity isn’t transformative — it is. But I think we need to be honest about the cost side of the ledger, not just the output side.
The MIT researchers called their study “Your Brain on ChatGPT.” They knew exactly what they were referencing. The question isn’t whether AI is changing how our brains work. It’s whether we’re paying attention to how it’s changing them — and whether we’re okay with the terms.
Any questions?
Update, Feb. 16, 2026: This article has been revised for clarity and framing.