The 5 skills every AI-era engineer needs

Every few months someone declares a skill obsolete. Most of the time they're wrong about which one. The engineers pulling ahead right now aren't abandoning the fundamentals — they're doubling down on the five that compound when you put a probabilistic tool in the loop.

1. Specifying problems precisely

When the model will cheerfully build whatever you ask for, the bottleneck becomes asking for the right thing. The ability to turn a vague want into a crisp, testable spec is now the highest-leverage skill on the team. It always was — it's just no longer optional.

2. Evaluation

If you can't measure whether an output is good, you can't safely automate producing it. Writing evals — defining "good," building a set of cases, scoring against them — is the skill that separates people who ship AI features from people who demo them.

3. Reading and reviewing fast

You'll generate more code, drafts, and options than ever before. The constraint shifts from production to judgment: reading critically, spotting the subtle bug, knowing what to keep. Code review was always a senior skill. Now it's the main event.

4. Systems and failure thinking

Models fail in unfamiliar ways — confidently, intermittently, expensively. Engineers who instinctively ask "how does this break, and what happens when it does?" build things that survive contact with real users. Those who don't ship impressive demos that fall over.

5. Communicating decisions

As more of the doing is shared with machines and teammates, the value of clearly explaining why goes up. Writing a decision down so others can trust it, challenge it, or build on it is no longer a soft skill — it's core engineering.

Notice the through-line: none of these are about a specific model or tool. They're about judgment. Tools will keep changing. Judgment is the thing that keeps paying.