scriptease.dev

Non-Load-Bearing

I was watching Matt Pocock's "AI Coding is exhausting," and thirty seconds in he hit a nerve I'd been ignoring for weeks.

His point was small and exact. The AI hasn't made the work harder. It's eaten the easy part — and the easy part, it turns out, was doing something.

Strategic vs tactical decisions

Here's the thing I couldn't explain to myself. I finish some days wrung out, and I've barely typed a line of code.

Matt reasons that work has two halves. There are strategic decisions — is this the right approach, is it safe, is it the shape I want — and there are tactical ones: the actual making, this line and then the next, the small moves that turn an idea into a working thing.

Strategic is the draining kind. It's judgment all the way down, and it never lets up.

Tactical is the other kind — the reason most of us started doing this in the first place. It's where you build something out of nothing with your own hands, and it feels less like deciding than like making. Still work. Just the good kind.

The AI is brilliant at the tactical half. So now my day is only the draining one. I got promoted to lead developer without asking, and so did everyone else — a full day of calls you can't afford to get wrong, with none of the making in between.

The making was the part I miss.

The break that got automated

Losing the tactical half didn't only cost me the fun. It cost me the rest.

The making doubled as a break. Low stakes, hands busy, head loose — it sat between the hard decisions like a landing between flights of stairs. You don't notice a landing is recovery until it's gone.

Now it's stairs the whole way down.

So I started clawing some of it back. I tell it to pull this out into its own function with better names, swap in a better library for the job, change the flow to be more readable, leave a comment about why we chose X over Y.

None of it is busywork. Every one is a real call, and the AI agrees every time — yes, better, good catch.

But being agreed with isn't the point. The decisions are small, and they're mine. A dozen of them is what keeps the code still feeling like mine.

Whose code is it

Because that's the quieter half of the exhaustion, the half Matt didn't have time for in thirty seconds. When a machine writes it and you only grade it, none of it feels like yours. And work that doesn't feel like yours is heavier to carry, even when there's less of it.

But the machine writing the code doesn't actually take the ownership away. I decided that. It's still my code if I put a hand on it — one choice that's mine, a shape only I would have picked.

This blog is the same trick. A machine helped write these sentences. They're still mine, because the story is mine and the voice is mine, and it only ever writes a line I would have said. The help isn't the opposite of ownership. Skipping the personal touch is.

I felt great after watering tomatoes

The real discovery wasn't at work. It was at home, on the evenings that put me back together.

They weren't the evenings I spent away from the AI. They were the ones I spent aiming it at something gloriously unimportant.

Teaching it to water my tomatoes from my phone — a cheap WiFi pump, a Royal Gardineer NX-4985, and one typed word that dribbles thirty-five litres onto the planters. Getting a reminder to tell me which bin goes out the night before, so I stop running down in my socks after the truck. Wiring up the robot vacuum. Writing a post like this one.

I thought it was just hobby time. Then I saw the pattern, and it embarrassed me a little, because it was the same thing I do all day at work.

Non-load-bearing

It's the same act. Point an AI at a problem, make call after call, steer it to something that works. That act flattens me by 6pm. At the kitchen table it recharges me. The only thing that changed is what's riding on the answer.

At work, every decision is load-bearing. Get it wrong and a customer feels it, or a colleague does, or a release does. The weight is the whole point of the job.

My tomatoes are not load-bearing. If the pump misfires, a plant gets thirsty. Nobody reviews my watering schedule. If the bin reminder is a day off, I laugh and take out the recycling anyway.

So my judgment gets to idle. I'm still deciding, still building, still in the loop with my hands on it — but nothing is at stake, so none of it costs me anything. It's the landing between the stairs, rebuilt out of play. The tactical break I lost at work, handed back to me by the same tool that took it, pointed the other way.

Load-bearing or non-load-bearing — you decide

Except the line won't hold still. I know because I watched one tool cross it four times.

I once wrote a little script — usage-check, a fuel gauge for my daily AI — showing how much I'd burned across both my subscriptions, side by side, its numbers folded into a briefing I read each morning. Both in one place, just to watch the numbers go up. If it stopped working, nothing was lost — it waited until I picked it up again.

Then a new model arrived that could fan out into a whole crew of helpers at once, and I could burn a day's usage before lunch. My toy became a gauge I steered by — and then more than that. It learned to stop the AI when I hit the limit and start it again when the five-hour window reset, so the work could carry on without me. Once it sailed twenty-eight hours straight. Load-bearing.

Then the limits themselves changed, the gauge broke, and I was sailing blind. So I became the manual usage-check — eyeballing my own numbers and halting the AI by hand, one word: pause. Load-bearing, and urgent. I fixed it in the middle of the night to keep the AI running.

A week later the same script reported, in that morning briefing, that the gauge had broken for the other AI — the non-load-bearing one. I fixed that too, but nothing rode on it this time, so it was the fun kind of work again. And while I was in there, I taught it to show the exact token counts, which I do not need in the slightest and which make the thing a small pleasure to read.

Same script — load-bearing and a toy at once, depending which half I looked at. Toy, then gauge, then emergency, then toy again. The code barely moved. What moved was what was resting on the answer. It's on version five now, so it has seen some things.

Pick something that doesn't matter

The trick runs backwards. Don't step away from the AI when it's worn you out — point it at something where nothing rides on the answer.

It doesn't even have to be its own project. It can be the useless corner of a serious one — the token counter I didn't need. What heals you isn't the tool; it's that nothing depends on getting it right.

Find the small, dumb, personal annoyance in your life — the plant, the bins, the light you keep forgetting, the spreadsheet nobody asked for. Give it an evening. Not to be productive. To feel the difference between deciding and playing.

The people who study this call it decision fatigue, and their fix is to protect your "craft time." This is craft time you'll actually take, because it's fun and it's yours and nothing depends on it.

The load-bearing hours will still be there tomorrow. The tomatoes needed me tonight.