AI Isn't Burning Us Out. The Narrative Around AI Is

Simon Willison says he can fire up four agents in parallel and by 11am he is wiped out for the day. Engineers at every major shop are publishing essays about AI fatigue. Business Insider is running surveys. SaaStr is forecasting a "2027 AI burnout wave."

The diagnosis is everywhere. The cure is nowhere.

I run a frontend team. I have AI woven through my entire workflow — in the IDE, in code review, in documentation, in research. By every metric circulating right now, I should be burned out. I am not.

So I started asking myself why. And I think the answer is simpler than the discourse makes it look: it is not the model that exhausts us. It is the story we keep telling ourselves about the model.

Myth one: AI is going to replace you

"AI will replace developers." "AI will take your job." "If you cannot 10x your output by end of quarter, you are already too slow."

You hear this every day. In the press, on LinkedIn, in comment threads under every post about software. Open your feed with your morning coffee and you already know you are behind.

That is what is tiring. Not the tool. Not the model. The relentless reminder that you have to run faster, or you will be left behind.

I know money is behind the hype. Funding, investors, courses, conferences, VC narratives — everyone has an interest in keeping things loud. But loud is not the same as true.

Maybe one day AI will replace certain professions. Maybe. I do not know. Nobody really knows. But that "maybe" should not be allowed to organize my present.

Myth two: more tokens equals better developer

This one bothers me more, because it is sneakier. It dresses up as productivity.

"The more tokens you burn, the more effective you are."

No. That is a myth.

Working with AI is not a slot machine. Enter, enter, take the code, ship it. That is not how anything good gets built. That is how you build a stack of things that nobody wants to refactor on Friday night.

Measuring AI work by tokens burned is the same mistake as measuring engineers by lines of code. Anyone who has shipped software for more than six months knows this. Sometimes your best commit of the day is the one where you deleted a hundred lines, not the one where you added a thousand.

It works the same way with AI. Your best session is not the one where you generated 5,000 lines. It is the one where you understood the problem and chose the right direction.

What AI actually is, if not a slot machine

AI is a conversation.

You iterate inside a session. You ask, you challenge, you refine, you reject. You step back and say "wait, are you sure?" Sometimes the model catches something you missed. Sometimes you catch something it does not know. That is intellectual work — except now you have a partner doing it with you.

That is the exact opposite of a slot machine, where you pull a lever and wait for an output. A slot machine does not think. A conversation does.

Slot machine	Conversation
Enter, enter, done	Question, refine, decide
Measured in tokens	Measured in understanding
Blame always lands on AI	Credit and blame are shared
"It generated this for me"	"I worked through it with AI"
No context	Context built together
Accept the first output	Reject it if it does not fit

The difference between those two columns is not the tool. It is how you treat the tool. The same Cursor, the same Claude Code, the same Copilot — one person uses it like a slot machine, another uses it like a partner. The outputs are different. The fatigue is different too.

Five rules I use as a Tech Lead

I am a Tech Lead. AI lives in every part of my team's workflow. Precisely because I see it up close every day, I rely on five rules I follow myself and share with the team.

1. Slow down before you speed up. The first message to AI should always be a question. Not "do this." Not "generate that." "Understand the problem with me." Five minutes spent describing context buys back an hour of not writing the wrong thing.

2. Challenge the first answer. The first output from a model is almost never the best one. It is the statistical average, not your decision. Your decision is the second one, the third one, sometimes the fourth iteration in.

3. Know what you do not want. It is easier to say "do not solve it this way" than to specify "do it perfectly." Negative constraints are a sharper tool than positive instructions.

4. Do not measure yourself in tokens. If someone on the team tells me they burned a million tokens today, I ask whether they learned anything. Tokens are a cost, not a measure of value. Electricity from the wall costs money too, but nobody brags about how many kilowatts they used this week.

5. When tired, write shorter. If a session with AI starts to drain you, that is a signal you slid into slot-machine mode. Stop. Close the IDE for ten minutes. Come back with one specific question. AI fatigue almost always shows up when you stopped thinking and started clicking.

What changes when you treat AI like a conversation

This is not abstract. I see concrete differences every day.

Better ideas — because you have someone to talk through a problem with at 2am when the team is asleep. Not finished solutions. The friction of bouncing thoughts off something that has no fatigue and no ego.

Faster work — because someone else takes the boring boilerplate. Not because you cannot write it. Because you do not want to write it for the 350th time. AI does not have that problem.

More headspace — for the things that actually need a human. Architectural decisions. Project context the model cannot see. Team politics. Business trade-offs. These still belong to you — and precisely because AI absorbs the noise, you have more room for them.

This is all there for the taking. But only if you stop treating every "AI will replace X" headline as gospel. And stop measuring your worth by tokens burned.

The problem is not the tool

Let me say this directly: the problem is not that AI exists.

The problem is that we let ourselves be carried by narratives we do not actually believe — because everyone else repeats them.

If you read ten posts about AI replacing developers, the eleventh one will tire you regardless of whether you agree with it. That is how the brain works. It is not a weakness — it is physiology.

And that is good news. Because if the problem is a narrative, the answer is a different narrative. You do not need meditation. You do not need a digital detox. You just need to stop buying the story being sold to you — and start buying your own.

My story is this: AI is a colleague. Not a boss saying "faster." Not a competitor saying "goodbye." A colleague who hands you the tool when you need it, and leaves you alone when you do not.

You do not have to run.

You can walk.

And you will get further.