I’ve been a developer for a long time. Long enough that I can sit at a keyboard for eight or ten hours, solving reasonably complicated problems, and not feel destroyed at the end of the day. It wasn’t always like that. Early in my career, I’d come home mentally wiped after a few hours. But years of doing it built something — call it muscle memory, call it automaticity — where the basic mechanics of reading code, holding a mental model, navigating a codebase, and typing out solutions became second nature. The hard parts are still hard, but the scaffolding around them is effortless.

Here’s the thing I’ve been thinking about lately: that muscle memory is highly specific.

The Programmer Who Can’t Read Documents

Take a developer who’s been coding for fifteen years. They can stare at a complex system, trace data flows across services, debug race conditions, and not break a sweat. Now ask them to spend a full day doing what an intelligence analyst does — reading through volumes of reports, cross-referencing sources, sifting through noise to find the one signal that matters, holding dozens of conflicting narratives in their head at once.

They’ll be exhausted by lunchtime.

Not because they lack the intelligence. Not because the task is objectively harder. But because they haven’t spent years building the cognitive pathways for that particular kind of work. Intelligence analysts train for years before they can sustain that level of reading and pattern-matching across documents for a full working day. Their brains have optimised for it. A programmer’s hasn’t. Every step requires conscious effort.

It works the other way too. Put an analyst in front of a codebase and ask them to follow a code review. The information density is overwhelming when you don’t have the trained patterns to filter and process it automatically. The fatigue sets in within minutes — not because they’re not smart enough, but because they’re exercising the wrong muscles.

This Is Exactly What’s Happening with AI-Assisted Development

When I started using Claude Code seriously, I noticed something unexpected. I was getting tired faster. Not physically — mentally. Which made no sense on the surface, because the tool was doing a lot of the work I’d normally do by hand.

But the work had changed shape.

Traditional development, the kind I’d spent years training for, is mostly about writing. You read some code, form a plan, and type out the solution. Your fingers know the patterns. Your eyes know where to look. It’s a well-worn groove.

AI-assisted development is more like directing. You’re specifying intent. Evaluating generated output. Deciding what to accept, what to reject, what to steer differently. You’re doing more reading than writing. More reviewing than creating. More high-level reasoning about what you actually want, and less low-level execution of how to get there.

That’s a fundamentally different cognitive mode. And my brain hadn’t built the automatic pathways for it yet.

The Fatigue Nobody Warns You About

I think a lot of developers try AI tools, feel this unfamiliar exhaustion, and conclude that the tools aren’t worth it. Or that they’re somehow slower with AI than without. And in the short term, they might be right — not because the tools are bad, but because they’re exercising cognitive muscles they haven’t developed.

Think about what’s actually different:

Specification over implementation. Instead of just fixing a bug, you need to describe the bug precisely enough that someone (something) else can fix it. That precision of description is a skill. It’s related to programming but it’s not the same skill.

Evaluation at speed. When AI generates fifty lines of code in two seconds, you need to read and assess it quickly. That’s a different reading mode than slowly building up code yourself. You’re scanning for correctness rather than constructing from scratch.

Letting go of control. This one’s subtle. After years of knowing exactly what every line does because you wrote it, accepting code you didn’t write requires a different kind of trust and verification. It’s uncomfortable until it isn’t.

None of these are impossibly hard. But they’re all slightly different from what traditional development trained us for. And “slightly different across many dimensions” adds up to genuine fatigue.

Why This Might Take Years

Here’s the part that I think people underestimate.

Traditional development didn’t become effortless overnight. It took years of daily practice before the basic patterns became automatic. Years before you could hold a complex system in your head without strain. Years before debugging felt like second nature rather than a puzzle you had to consciously solve each time.

Why would AI-assisted development be any different?

The developers I know who are genuinely productive with AI tools have been using them daily for a year or more. They’ve built their own workflows, their own patterns for prompting, their own sense of when to let the AI run versus when to do something by hand. That fluency didn’t come from a weekend workshop. It came from the same slow accumulation of practice that made them good at traditional development.

And we’re still early. The tools are changing fast, which means the muscle memory is a moving target. It’s like learning to code when languages and frameworks were evolving rapidly — you build competence, the landscape shifts, you adapt, you build again. Eventually the rate of change settles and things become more stable. We’re not there yet.

The Transition Is Worth It

I don’t want this to sound discouraging. The transition is real, the fatigue is real, but the payoff is also real.

Once you start building those new cognitive pathways — once “describe the problem, evaluate the solution, iterate” starts becoming as natural as “read the code, form a plan, type it out” — the leverage is enormous. You’re operating at a higher level of abstraction. Problems that would have taken a day take an hour. Not because you’re smarter, but because you’ve developed the muscle memory for a more powerful way of working.

But you have to put in the hours to get there. And you have to accept that the early hours will feel harder, not easier. That’s not a sign that AI tools don’t work. It’s a sign that you’re building new skills.

Give it time. Your brain will catch up.

Written February 2026.