Everyone is burning out on AI, and I think I know why it skips me

Something strange happened in my reading list this month. Three different people, none of them lightweight, all landed on the same note within days of each other: AI is wearing them out.

Evil Martians wrote that AI-assisted engineers are burning out and asked, half-seriously, whether that is fine. Their point is sharp: you ship more, but the joy of writing code gets replaced by the grind of reviewing code you did not write, and you get maybe four or five intense hours before your brain is fully cooked. Harvard Business Review ran a piece on "brain fry", built around watching an orchestration platform run a swarm of agents faster than any human can follow, with one user describing a real, physical stress just from watching the thing move. And Phil Nash built a small app with an agent and wrote that he genuinely does not know how he feels about it, because he did not write the thirteen hundred lines and is not sure the result is really his.

I read all three back to back and my first reaction was not "yes, exactly." It was: huh. I have been running harness engineering as my main way of working for over eight months now. The speed keeps going up. The volume of pages I read and decisions I take keeps going up. And I am not fried. I am not having the brain fry. I am, if anything, more energized than I have been in years.

So this is the puzzle I want to sit with. Not "they are wrong and I am right." They are clearly describing something real, and a lot of smart people are feeling it at the same time, which usually means it is not noise. The interesting question is the other one: why is it hitting them, and why is it skipping me? I have a few theories, and I think most of them come down to who you were before the harness showed up.

The wave arrived later than I expected, and then all at once

Here is the part that surprised me first. When this way of working clicked for me, I honestly expected a stampede. I thought half the industry would be on this train within weeks. Instead it stayed quiet for a while, and then a few months in, the mood flipped, and the posts I started seeing were not "look what I shipped," they were "I am exhausted and I do not know if this is worth it."

That timing told me something. The burnout is not arriving at month one, when everything is shiny and the demos feel like magic. It is arriving around month three or four, once the novelty is gone and what is left is the actual shape of the work: long sessions of oversight, constant context switching, reviewing output you have to trust and verify at the same time, and a finish line that keeps moving because the friction to try one more thing dropped to almost zero.

And that shape, the one that is breaking people around the four month mark, is not new to me. It is just Tuesday. It is the work I have been doing for almost twenty years under different names.

The profile thing nobody wants to say out loud

I think the biggest factor is profile, and I want to say this carefully because it is easy to make it sound like a flex, and it is not meant to be one.

Since day one I have worn a stack of hats at the same time: developer, project manager, DevOps, a fair bit of cybersecurity, client relations, the lot. I run a small company. There was never a version of my day that was "sit down and write code in flow for six hours." My day was always: switch context, make a decision, catch a glitch someone else missed, hold the whole system in my head, drop into the weeds for twenty minutes, pop back out to the architecture, answer a client, patch a gap. For me that is not a cost. After two decades it is a trained muscle. It is closer to a sport than a strain.

Now look at what the harness actually asks of you. It asks you to stop being the person in flow writing the lines, and to become the person overseeing, deciding, catching, steering, and signing off. It asks you to switch attention constantly and to deep-focus on demand. It asks you to find the one weird thing in a wall of generated output. For a developer whose whole identity and whole reward system was built around the craft of writing the code themselves, that is a brutal trade. You took away the part they loved and replaced it with the part that, for them, was always the tax. Of course that burns. The piece Evil Martians wrote nails this without quite naming it: the fulfillment moved, and for a lot of people it moved somewhere they do not enjoy being.

For me almost nothing moved. The harness just handed the redundant code-monkey work to the agents, the part I never identified with anyway, and left me with the overseeing and the fine patching, the part I was already doing for everyone else on every project. Phil's worry, "I did not write it, is it even mine," does not bite for me the same way, because I stopped measuring my output in lines of code a long time ago. I am not coding it. I am making sure it matches my expectations and my specs, and that has been my actual job for years. The model writes. I decide whether what it wrote is allowed to exist. That feels like ownership to me, the same way a building is the architect's even though they did not pour the concrete.

Reward is the thing that protects you, and mine is huge right now

Burnout is not just hours. It is hours without reward. You can do absolutely punishing work for a long time if every push gives you something back. You break fast doing easy work that gives you nothing.

My reward right now is almost embarrassingly large. I have a shelf, and on that shelf are projects I have been carrying around for years. Some of them have been sitting there for close to a decade, because as a single person the cost of laying down all the boring foundation, the redundant plumbing, the scaffolding, was always more time than I had between client work. So they waited. For years they just waited.

That shelf is emptying now. I am finally launching things I genuinely wanted to exist in the world, and I get to do it in a way where I can actually build maintenance, security, and scalability in from the start, cement them, and keep improving them over the long run instead of bolting them on in a panic later. That is not exhausting. That is the most rewarding stretch of work I have had in a very long time. When the reward is "the idea I have been sitting on since 2008 is live and people can use it," four hard hours of review feels like nothing. It feels like a gift.

I think this is the quiet variable under a lot of the burnout stories. If you are using the harness to ship more of the same work for the same employer at a faster pace, the reward did not scale with the intensity. The intensity went up and the meaning stayed flat. That gap is where people are getting cooked.

One stack, held for years, so the harness compounds

Here is the concrete, unglamorous part, and I think it matters more than it looks.

I run the same stack for almost everything: Nuxt 4 with Nitro on the server, and then SQLite, Postgres, or Mongo underneath depending on what the project actually needs. That is it. It is extensible enough to carry basically any kind of project I take on, and I have been refining how I build on it since long before agents were good.

What that buys me is not just familiarity. It is that every hour I spend improving my harness, the tests, the spec-driven flow, the product architecture conventions, the database patterns, the way I lean on what a modern framework like Nuxt and Nitro already gives me, that hour pays off on the next project, and the one after that. The harness compounds because the ground under it never shifts. I already know every common Nuxt, Nitro, and Vue pitfall by heart, so I can write the rules and the guardrails that keep the model out of the ditch, and they keep working.

Compare that to the churn the burnout posts are full of. A new model every week, a new orchestration tool, a new "you have to try this," and underneath it people who are also still arguing about whether it should be React or something else this quarter. I do not have any of that. I do not lose a day to a framework debate. I do not lose an afternoon to "should we switch." There are no meetings to justify the stack, no brainstorming sessions that end where they started, no relitigating decisions I settled years ago. The tools just work, because I made them work for one stack and then stopped touching that question. A huge amount of the fatigue I read about is not the AI at all. It is the constant re-deciding of things that should have been decided once. I wrote about this churn before in my piece on AI FOMO and cognitive overload, and it is still the thing I see draining people the fastest.

Why this is so much harder inside a big team

This is the part where I want to be most careful, because it is easy to turn it into a dunk and I do not mean it as one.

A lot of the people burning out are not burning out because they are weaker. They are burning out because they are inside structures that make harness engineering genuinely miserable to adopt. A solo operator or a small, tight team can re-shape the entire way of working overnight. There is nobody to convince. I can decide on Sunday that the process changes, and on Monday it has changed.

A big team cannot do that, and not because the people are bad. There is real friction baked into the building. There is honest fear, the very reasonable worry that this tooling exists to replace the person being asked to adopt it, and you cannot expect someone to enthusiastically optimize themselves out of a job. That fear curdles into quiet resistance, sometimes into outright internal sabotage, and the process that comes out the other end is bloated, personalized, half-owned, and impossible to put under one roof. So the individual engineer ends up carrying all the intensity of the new way of working with none of the clean autonomy that makes it bearable. That is a structural trap, not a personal failing, and I have a lot of sympathy for the people stuck in it.

A few years ago this same consistency and focus already let me outrun agencies and teams much bigger than me. The harness widens that gap, because the part I am good at, a clear process aimed at a focused goal, is exactly the part that can be iterated, optimized, and automated over and over. The thing big teams struggle to apply cleanly is the thing a single aligned person can sharpen endlessly.

I am not made of stone, and I want to be honest about that

I do not want to pretend I am immune, because that would be both untrue and kind of insulting to the people who are struggling.

I still hit hard limits. I burn through token budgets because I always want to use everything I am paying for and I always have more ideas queued than the day can hold. And yes, auditing thousands of lines, being the overseer instead of the author for a long stretch, is the single most tiring part of all of this. On a single product it gets monotonous. Across several products and stacks it can drain you. I cannot do it endlessly and I do not pretend to. There are still parts that need real excellence and a different mindset, the parts where the model genuinely cannot step yet, and those are mine to carry by hand.

So my honest position, and I will put the caveat right in the open, is this: at least while I am still not burned out, harness engineering has been one of the best things to happen to how I work. That "at least while" is doing real work in that sentence. Maybe it comes for me too. Maybe in another eight months I write the tired post. I am not betting on it, but I am not arrogant enough to rule it out.

Two sides of one coin

The thing I keep coming back to is that the people writing these burnout pieces and I are looking at the exact same tool and having opposite experiences, and both of us are telling the truth.

The harness is a genuine gift for some people. For a multi-hat veteran with one well-worn stack, a strong sense of product, and a shelf full of things they have always wanted to ship, it removes the exact friction that was holding them back and leaves them doing the part they were already best at. For other people, people whose joy was in the craft of the code, or who are trapped inside an org that cannot adapt cleanly, or who are shipping more of the same with no extra reward, the same tool strips out the satisfying part and hands back the tax. That is not a flaw in them. The coin just landed on the other side.

We are all unique, which is the whole point, so it would be strange if a tool this big felt the same to everyone. This is only my take, from my seat, and I am sharing it because the loudest version of the story right now is the burnout one, and I think the other side deserves an honest voice too. Not a brag. Just the other half of the coin.