Stop hiring “AI Engineers”
The title describes a tool, not a job. Hire for the failure mode you actually have, and the req writes itself.
I have read a lot of job descriptions for AI Engineer, and I could not tell you what most of them are asking for. The title describes a tool. Hiring for a tool is how you end up with five people who all know the same API and nobody who can tell you why the output got worse.
Hire for the failure mode
Write down the way your AI work currently fails, in one sentence, and the req writes itself. If your problem is that quality moves and nobody notices, you want someone who has built measurement into a system before. If your problem is that inference costs are compounding faster than usage, you want someone who thinks in unit economics and is comfortable saying no to a bigger model.
The two-question filter
In an interview I ask for something that shipped and got worse, and what they did about it. Then I ask what it cost. Candidates who have carried a system through its second bad quarter answer both without preamble; the rest tell me about a demo.
None of this is free. Hiring for a failure mode means writing a different req for each opening, which is slower, and it means you cannot hand recruiting a single template and walk away. You also give up the option of hiring in bulk against a title the market recognises, which is a real cost when you are trying to grow a team fast.
What you get back is a team where two people are not doing the same job, and where the person who owns the number is the person who can move it.
What is the sentence describing how your AI work fails right now?