The boilerplate is getting free. The scarce skill is deciding what the product is supposed to do — and making sure the agent obeys that intent.
The low-level code that makes a product go through is getting cheap, fast. An agent can scaffold auth, wire a form, and ship a checkout in an afternoon. That used to be the expensive part. It isn't anymore.
What is becoming valuable is the judgment above it: knowing what the money, the user, and the product are supposed to do — and making sure the generated system obeys that intent.
The new split in product work
I keep coming back to the same split. Agents are excellent at producing plausible surfaces. They are weaker at holding a commercial thesis, a constraint, or a taste bar across a multi-day build. That gap is the product work now.
So the practice shifts. Spend less time typing the first draft. Spend more time writing the signals the draft must satisfy.
- Who it's for — and who it isn't
- What success looks like in one sentence
- What must never happen
- How you'll know when the agent drifted
Judgment as the product layer
When code is cheap, taste and constraint become the real interface. The builders who win won't be the ones who generate the most. They'll be the ones who can steer generation toward something people actually want to use — and pay for.
What I'm doing differently
I write the judgment first: intent, constraints, acceptance checks. Then I let the agent draft. Then I review against those signals, not against whether the demo looks impressive.