Level Up the Game

I was scrolling LinkedIn when a post from Kent Beck stopped me. Beck gave us Extreme Programming, and these days he writes about coding with AI as openly as anyone I follow. This post opened on him asking for his struggle back, because he needs it, and the thing taking it is the genie, his word for AI.

My first thought came out before I'd finished reading, and it came out as three short lines:

Level up the game. Go after the impossible. Seek the inconceivable.

This is me working out what I meant.

What Beck is asking for

He starts with an image I wish I'd come up with: a gym where a machine does the lifting. You press a button and watch a weight go up and down, you set how heavy it is, and a counter tells you how much energy your workout burned. None of that is a workout. His worry is that genies are being built the same way for the mind: press, watch the output, feel productive.

Then he splits the work in two. Toil is work where doing it teaches you nothing. He had the genie reformat an old essay of his that had lost its formatting, and doing it by hand would have left him the same person with less time. Struggle is the other kind. His example is a freshman-year train ride, stuck for hours on a theorem he knew was true and couldn't prove, until it gave. What he came away with was more than the proof. He found out he could take on problems harder than anything school had given him. It's a lovely story. He ends the LinkedIn version with the line I keep thinking about:

"Let the genie carry the toil. Let me keep my struggle."

The longer version is in his newsletter, Take My Toil, Leave Me My Struggle, and it's worth reading in full.

I agree with most of it. His toil and struggle are the same split I wrote about in Germane Friction, the effort that only taxes you and the effort that builds you. I've felt the gym too. I've handed a piece of thinking to an agent because it was hard, gotten the answer back, and known that I skipped the rep.

He tells them apart by what the work leaves in you once it's done, and I think that's exactly the right test. It's also why I see the ask a little differently. Near the top of the newsletter he wonders what genies would look like if they weren't built as a RoboGym. My guess is that the genie can't do the sorting for me. It can't see what a task would have left in me, and the same task is toil for one person and struggle for another. So I don't expect the genie to leave my struggle alone. I expect to go and find it.

And he points that way himself. The prize on the train wasn't the theorem. It was finding out he could go harder. He ends the newsletter on the same note: the learning and growing you do today are what set you up for bigger goals tomorrow. That's the part I want to push on, because with a genie beside me, some of tomorrow's bigger goals are within reach today.

Move the target

The genie takes the struggle out of the wishes I already had. That's what it's good at. The function I used to write by hand, the config I used to look up, the draft I used to grind through. Those wishes were sized for a person working alone, and for a person with a genie they're too small.

So make the wish bigger. Aim at something I couldn't have done alone, and couldn't have done with the tools a year ago either, and the struggle doesn't need protecting. It's right there at the new target, waiting, and no genie removes it, because the genie doesn't know how to do it either.

The struggle didn't go anywhere. It moved to the edge of what you're trying to do.

That's what I meant by level up the game: leave the old level instead of grinding it harder.

What it looked like

I've been doing this without calling it anything.

This month I set out to read the founding papers of cybernetics, Ashby and Wiener and the rest, on my own. That was already past where I'd normally go. I opened the first paper and bounced. An agent can check a quote for me. It can't read for me. So I built a map first, a companion and a primer for every work on the list, and then I still had to sit with the papers, and some of them were still steep. Without the genie I wouldn't have tried, and with it the reading was still mine to do.

Then I tried to watch one of Ashby's laws do work in a real system, a reconciliation loop on Kubernetes. The loop was wrong twice before it was evidence, and at the end the law turned out not to apply the way I expected. Nobody could have handed me that. The struggle was the result.

Most of what I've done since I got into engineering in collaboration with AI looks like this. It has pushed me much harder, to do more and learn more, and to prove things out. It's a different way of learning and working, a new paradigm. I wrote a book about it while it was happening, AgentSpek. Its chapter on the knowledge spiral starts from the idea that a new tool doesn't solve your problems, it frees you for other ones, and the other ones turn out to be harder. The idea was in the book before I had the words for it.

Impossible, inconceivable

I don't mean impossible like physics. I mean not possible for me, yet. Things I wouldn't have started, because the distance from here to there was too far to walk alone.

Inconceivable is the next step out. It's the thing I couldn't have pictured wanting before, because there was no road to it. When the genie makes the old wish list trivial, the list has to grow, and some of what goes on it I couldn't have imagined before. That's the part I find exciting. It's also where most of the hard thinking is now.

None of this works if I let the genie do the reaching. It can carry me further out, but the judgment out there is still mine to build. That's the same discrimination Germane Friction is about, done at a harder target. And it's the same thing I wrote in The Right to Explore: not understanding yet is where the learning starts, as long as the loop closes against something real.

Keep what it teaches

One more piece, and it's the one I'd skip if I weren't careful. The hard session at the edge is where I learn the most, and it's also the easiest session to walk away from with nothing. The thing works, I'm tired, I close the laptop. What broke, what the model got wrong, what I finally understand: gone by morning.

So I write it down. My handwriting goes through two models and I correct them, and every batch ends with the misreads added to a lessons file the next batch starts from. That file is small and it gets better every time. It's the part of the work I'd least like to lose.

If I'm going to go after the impossible, I want to come back from it knowing more than when I left.

Back to the gym

If the genie is a machine that lifts the weight, I don't want to unplug it. I want to put on a weight the machine can't lift by itself, and get under it with it.

I think Beck and I want the same thing, the struggle that makes you better at the work. I'd just rather go out and find it than ask the genie to leave it where it was.

Related reading: Germane Friction, the two kinds of effort and why the tool can't tell them apart. Working at the Frontier, what it costs the body to work out at the edge. AgentSpek, the book, from the first overwhelmed week to becoming someone who works this way.