If an engineer is not AI-native, their judgment can drift at the level of cost estimation.
Engineering judgment is calibrated by a default cost model. System design, risk sense, engineering taste — all of them eventually have to answer two questions: is this worth doing, and how expensive is the remaining work?
Coding agents changed implementation cost. Prototyping, debugging, refactoring, validation — all of them got cheaper. Once the cost structure changes, many previously reasonable effort estimates start to move away from reality.
A feature that takes a week in a human-only workflow may have most of its real problems solved in half a day in an agent-native workflow. The remaining work looks like the same work. The actual cost is different.
Many team decisions depend on a cost model. Whether to descope, keep polishing, refactor, or build an initial version — these are judgments about cost, value, risk, and timing.
If the cost model has not updated, these decisions become systematically conservative. A feature close to completion can look still too large. A direction that could be quickly prototyped can die in discussion. A problem that could be aligned through an artifact can stay stuck in abstract scope debate.
There is a subtle friction here: experience itself can become resistance.
I do not think age is a good proxy for whether someone is AI-native. That judgment is too crude. But cost-estimation errors are more likely to show up in experienced engineers, because experience contains a lot of intuition formed under an older cost model: what is hard, what is expensive, what is not worth doing.
Newer engineers may have less of this problem. They have fewer memories from the old workflow, so they are less likely to price current work with past costs.
P.S. Welcome myself back to the blog. Cloud Next + I/O 就不能隔远点吗.