Writing software used to be slow. Because it was slow, companies optimized for patience. They hired people who could hold a large system in their head for a long time, and they planned in quarters. Speed was nice to have. What you really needed was someone who would not get lost.
That premium is gone. An engineer who works well with models does in a week what a team used to do in a quarter. The typing is no longer the hard part. The hard part is knowing what to build, noticing when the machine is confidently wrong, and having the taste to throw away the nine bad versions and keep the good one.
Most engineers are not set up for this. Not because they are not smart. The habit of a career is hard to drop. People who spent ten years learning to build slowly and carefully are still being paid to build slowly and carefully, and the reason to change is weak until it is sudden. A new kind of engineer learned the job after the models arrived. They never built the old way long enough for it to set. They reach for a model the way the last generation reached for a database. They are native to this, not retrofitted to it.
What does that actually look like. It is not someone who pastes prompts into a chat window and ships the first answer. It is someone who treats the model as one more part to engineer around. They build evals before they trust an output. They know when to hand work to the model and when the model will waste their afternoon. They read the failure cases instead of the happy path. They ship, watch what breaks, and fix the system rather than the symptom. The model made them faster. The judgment is still theirs.
These people are rare, and they are easy to miss. The thing that makes them good does not show up as a credential. It does not show up in years of experience, because the field is younger than the experience most resumes are sorted by. It shows up in how they work, which means you have to watch them work to see it. A normal hiring process is built to do the opposite. It filters on the signals that are easy to read and slow to change, which are exactly the signals that miss this.
So the people who can do this are spread thin across companies that do not know what they have, and the companies that need them cannot tell them apart from everyone else who says the same words on a resume. That is the whole problem. Demand is high, the supply exists, and the market cannot match the two, because the thing being traded is invisible to the usual tools.
This is the part we decided to do. We find these engineers and we learn how they work, not how they interview. We are slow about it on purpose. We would rather be sure about a few people than have an opinion about many. The bar is simple to state and hard to clear. We only hire the best AI-native engineers.
We think this is the biggest shift in how software gets built in a generation. Shifts like this do not announce themselves. They look like a few teams quietly shipping faster than they should be able to, until one day the gap is too large to explain any other way. We would rather be early to it than right about it later.