The Draft Is Not the Point

I've been working close to 18 hours a day. Not to boast or anything. That's just where things are right now.

With Claude Code, Codex, all these agents. Things I used to only dream about building, products, websites, all that stuff. They're just... happening. All of a sudden everything feels achievable. Everything feels fast.

And yet there's this hollow feeling I can't shake. A sense of achievement, no sense of achievement. Both at the same time. I don't know how to explain it better than that.

I think I've been measuring the wrong thing.

Think about how writers work. The ones who've figured it out sit down and write a terrible first draft, not because the draft is the point, but because it's how they get to the point. The draft is in service of the vision. The real work is knowing what you're trying to say. The draft is how you discover it.

Here's the thing though. The best software engineers were already doing this. Fred Brooks wrote in 1975: "Plan to throw one away. You will do that anyway." Kent Beck built spike solutions into XP in the 90s, code you write specifically to learn from and discard. The thinking was always there. But most teams never actually did it. The pressure, the sunk cost, the "we've already spent three weeks on this", that's what held everyone back.

What LLMs have changed is not whether you can throw code away. They've changed the cost of it being okay to throw it away. When a human wrote the code, there was weight to discarding it. Now there isn't. Which means you can finally work the way the best people always said you should: quick rough pass, understand what you've got, be ruthless, start again.

The mistake I was making is sitting in front of an agent for 18 hours thinking that's the work. Most of that time, the agent is building. What I'm actually contributing is the deciding. What to build, in which direction, why this and not that. And that kind of thinking, I want to be honest, it doesn't just happen on a walk. It needs the desk time first to load the problem properly, but it also needs space outside the screen. Most of us just don't count that as work because there's no output to show for it.

Your years of experience, your taste, your sense of what's right. That's what an LLM doesn't have. That's what guides it toward something genuinely good versus just something fast.

The shift I'm trying to make: less time grinding, more time knowing what I'm actually trying to build.

That's the whole thing.