No. 26AI Productivity tools

Don't spend the time you save

A friend sent me an article about over-optimisation, and I knew roughly what it would say before I opened it. Steven Strogatz and Alex Townsend, two Cornell maths professors writing in the Guardian, set out the traps of optimising everything: squeezing out all the slack, defining “best” too narrowly, and lots of individually sensible optimisers adding up to turmoil. AI, they argue, could make every one of them worse.

I agree. But the trap I keep coming back to is smaller and more personal. It’s what happens to your attention when you get very good at getting things done.

The reward for efficiency

Forget AI for a moment. Long before any of this, the most organised person in the office had a familiar problem. If you can do a job in half the time it takes someone else, you don’t get to go home early. You get given more work.

Now you’re doing twice as much as everyone else, with your attention split between two things. You get good at juggling those two, so you speed up again, and the reward is a third. The prize for efficiency was never rest. It was more work and a bit more exhaustion.

Psychologists have measured a version of this. People who come across as highly self-controlled get handed more of the load, and the people handing it over assume it costs them less effort than it does. The people themselves said it took them just as much effort.

Ten things underway

AI turns that dial right up. You can build yourself little helpers for your own email and calendar, reading what comes in and drafting replies for you to check. Some jobs just vanish. When I want screenshots of one of my little games for a build note, I used to load it up, get to each spot, take a screenshot, crop it, rename it, save it in the right folder and drop it onto a board. Easily an hour. Now it’s a thirty-second prompt, and it happens in the background.

Here’s the catch. The job doesn’t finish in thirty seconds. It runs while you start the next thing, and the next. You’re not working on ten things, but you’ve got ten things underway, and some days I’ve had something like ten times as much on the go as I used to.

That’s when you lose the thread. Results come back fast, they look good, and you start trusting the machine’s taste. You stop checking properly. You’re working ten times as fast, but the work isn’t ten times as good. Often it isn’t even as good as before.

The slow bits used to be where the checking happened. Doing it by hand, you looked at each piece as you went. Automate the doing and you’ve quietly automated away the looking. That’s how polished-looking rubbish gets made, the stuff researchers have started calling workslop. When BetterUp Labs and the Stanford Social Media Lab surveyed 1,150 US desk workers, four in ten had been sent some in the previous month.

Spend the saving on checking

So I’ve started pulling the brakes. Software teams have a name for this: a work-in-progress limit, a cap on how many things can be in flight at once. I’ve started setting one for myself. Even if I could spin up ten jobs, I’ll start a few, let them do their thing, and take my time checking what comes back.

Think of a job as ten units of time. Before AI, maybe seven went on building it and three on checking it. Now the building might take one. The tempting move is to spend the other nine starting nine more jobs. The better move is to spend most of them checking, trying it out and giving feedback. The job might take as long as it always did, but what comes out is better.

Fig. 1 · Where the saved time goes

Same job, same ten units. Building used to take seven of them and now takes one. The choice is what the other nine are for.

I’m all for automating the monotonous stuff. Making the folder structure and filing things in it under sensible names. Writing the covering email when everything it needs to say is already sitting there. Those jobs send me up the wall and don’t make me better at anything. Automate away.

What I don’t want to save time on is the thinking. The time freed up should go back into thinking, not get stolen by the next task.

Design dandruff

The other way the saved time gets wasted is adding things just because you can. With AI, anyone can bolt anything on. An old colleague of mine called it design dandruff, and the term stuck with me: all the little flourishes and extras a thing doesn’t need.

I think about a good product the way I think about a good piece of art. Add anything and it’s too much; take anything away and it’s not quite right. Saint-Exupéry said much the same, writing about aeroplane design as it happens: perfection arrives when there’s nothing left to take away. That holds for a video as much as for a game.

So the skill now is knowing what not to build. In the seven pillars I argued that skill has narrowed to taste, and this is what I meant. Taste is what tells you to leave something out, and the machines can’t hand it to you.

One good game

Here’s what that looks like for me right now. I’m making a little browser game where you fly a cartoon plane over the Nazca Lines in Peru, tracing the giant figures to colour them in. I had a playable version within a day, something that would once have taken me months. I put real effort into the engine, and a lot of the routine work, from the code and the deployment to digging out the shapes of the figures, was done with AI.

But the time saved isn’t going on more games. It’s going on playing this one: flying it to feel how the plane handles, and giving it loads and loads of feedback. I’m spending the saving on the thing that actually matters, which is whether it’s fun to play.

I don't want to make ten games because I can. I want to make one very good one.

I don’t want to make ten games because I can. I want to make one very good one.

For years the reward for being efficient was just more work. Now the machines can do so much of the doing, don’t hand the time straight back. Keep it for checking, and for a bit of breathing space.

Book links go to Bookshop.org, which pays me a small commission and gives the same again to independent bookshops.