Hire for AI fluency, not AI awareness

"Proficient with AI tools" now appears on a majority of the resumes we see, which means it has stopped carrying information. The gap between the people who wrote it because it's true and the people who wrote it because it's expected is enormous, and no keyword filter will find the edge of it.

Three levels, and only one of them matters

Aware. Has an account. Uses it for first drafts and the occasional summary. Would describe AI as "helpful." This is now the floor, and a floor is not a differentiator.

Fluent. Has AI in the daily workflow, not beside it. Knows which tasks to hand over and which to keep. Prompts iteratively rather than hoping. Reads output with suspicion, checks the parts that matter, and edits hard. Can tell you, unprompted, what the tool is bad at in their domain.

Compounding. Builds. Templates, chained prompts, small automations that outlive the task they were made for. Turns a recurring three-hour job into a twenty-minute one and then hands the twenty-minute version to the rest of the team.

Fluent is the hiring bar. Compounding is the ceiling you're hoping for.

Why interviews can't tell them apart

Ask "how do you use AI in your work?" and everyone has an answer, because everyone has read the same articles. The vocabulary is free. What isn't free is the judgment underneath it, and judgment doesn't surface in a description of judgment.

The only reliable test is to hand someone real work, let them use whatever they'd normally use, and read what comes back. Fluency is legible in an artifact within about ten minutes:

The aware candidate's submission is fluent-sounding and slightly generic, with a structure that matches whatever the model defaults to and constraints from the brief quietly missing. The fluent candidate's submission is specific, shaped around the actual scenario, and shorter than you expected, because the editing was the work.

What we ask for

During Day One™, AI use is expected and disclosure is required. Which tools, for which parts, and what you changed. Not for compliance theater: the disclosure is the assessment. Someone who can narrate their own leverage accurately understands it. Someone who can't, doesn't, and the narration is where that becomes visible.

The strongest answer we get sounds roughly like: drafted the client email with the model, rewrote the second paragraph because the tone was too soft for an escalation, did the calendar logic by hand because it kept inventing a conflict that wasn't there. That's a person who knows the edges of their tools.

The trap on the other side

Optimizing purely for AI output quality selects for people who are excellent at generating volume. Volume is not the constraint any more; taste is. The scarce skill is knowing which of five plausible outputs is right, and being willing to throw away work the machine produced convincingly.

So the thing to hire for isn't speed. It's speed plus refusal: how fast someone can produce, and how readily they'll discard.

What this means for how you write the role

Stop listing tools. Tool lists date within two quarters and they filter for people who collect logos. Describe the leverage you expect instead: we expect the first draft of anything to be machine-assisted and the final version to be yours; we expect recurring work to get faster every month you own it.

Then test for exactly that, and let the submission decide. It is the same reason every SupaHuman is vetted for AI fluency before they reach a client. The lowest use of a person is as an expensive API for repeatable tasks. The highest is judgment, taste, and knowing what matters, which is precisely the part the tools can't do, and precisely what you should be hiring.