Agent Coding Is Not Product Making

Lately, I’ve watched a familiar arc: someone has a great idea, dumps it into a coding agent, and a few days later has a working product that leads nowhere. There are exceptions, of course. But the pattern makes sense.

The idea gate used to be huge. Building was hard, and that forced you to decide whether an idea deserved the effort. You thought about feasibility, cost, market fit, competitors, technical challenges, how to reach people, and whatnot. An MVP was minimum because even the minimum was expensive.

Writing the code pushed back, too. You would find a corner case that broke the premise, discover that a competitor had solved the hard part surprisingly well, or become more confident that you were onto something. The friction was inconvenient. It was also research.

Agent coding removed much of that friction. Good. I use it every day. But it also made it easy to skip the thinking that the friction used to force.

You don’t ask whether anyone cares. You just do.

You don’t talk to a handful of people who might use it. You just do.

You don’t build in public, because someone might steal your idea. You just do.

You don’t pause a hundred times while coding to reconsider the product. You just do.

You just do.

The danger is not only making something nobody wants. You may never stay with the idea long enough to care about it, obsess over it, and make it good. You may not even notice that the idea is genuinely great because you are already rushing toward the next one.

Cheaper coding did not make product thinking cheaper. It only made it easier to avoid.

Agent coding can speed up coding. It cannot tell us what deserves to exist.

I’m not going to stop using it. Hell, no. I just don’t want a tool for moving faster to become an excuse for thinking less.