When hiring for senior roles, many companies use a project of some sort to assess candidates. Recently, we’ve noticed a trend toward companies prohibiting candidates from using AI on those projects.
(I’m picturing a job candidate sitting in a mostly empty room with a table, chair, pencil that needs sharpening, and one of those small, thin blue notebooks that schools use for testing.)
There’s probably a better way.
Let them use AI. But don’t just review the output. More importantly, review the transcript of the conversation the candidate had with the AI that produced the result.
AI is here to stay. On the job, we all use the tools available to us, and we should. Why would we make our job harder than it needs to be? Our goal is to produce a result. The highest quality and fastest result possible with the available tools. AI is one of those tools. We shouldn’t hide from it. We should embrace it.
So when we evaluate job candidates, we should measure who can do the best job with the tools they’ll actually have on the job. You’re hiring them plus AI, not just AI.
Here’s why the transcript matters. When I use AI for something I’m highly skilled at, my transcript looks completely different from when I use it for something I know nothing about.
In the first case, I’m both leading and following the AI to the answer. I’m giving good instructions, detailed prompts, vital context, important pointers, correcting the AI when it drifts, adding my own value, and acting as a true partner. In the second case, I’m still good at using AI, so I ask a lot of good questions. Plus, AI is still smart, so it will produce a pretty good result. But I’m more of a follower than a leader, and the result will be a much higher percentage AI versus me.
Here’s a simple example to illustrate: When I prompt AI to create an image, I’ll say something like, “Make a picture of a person standing on a street corner in NYC, with a lot of people around them staring at them.” An expert would say something like: “Street photography, eye-level shot with a 35mm lens, of a woman in her 30s in a camel coat standing still on a Midtown Manhattan corner at dusk. Dozens of commuters around her have stopped mid-stride and are staring at her. Shallow depth of field, so she’s sharp and the crowd is slightly soft. Light motion blur on the cars behind them. Warm storefront light mixed with blue hour sky. Her expression is calm, not alarmed. Shot in the style of a candid documentary photo, natural skin texture, no text or signage legible, 16:9”
And then when the first version comes back, I might say: “Make the person look happier.”
While the expert would say: “Make half of the crowd mid-step and a few holding phones. Her face is too symmetrical. Pull the camera back so we see more of the intersection. Change the lens to 24mm.”
You get the point.
The transcript shows you two things at once: how good someone is at using AI, and how much value they contributed to the result. Experts ask the best questions. They give direction. They know what’s missing. They push back. They steer the conversation toward the right answer instead of accepting the first one.
You can see that difference in the output, but it’s far more obvious in the transcript.
So, you can still use a project to assess a candidate. Skip the secure room, armed proctor, paper, and pencil. Let them use AI. Review the output. Then read the transcript. You’ll get an accurate read on how they’ll do the job, with the tools they’ll actually use to do it.
(You should see my AI transcript for creating this image. You would definitely hire me to be your image generator.)
