Most "do you use AI?" questions get the same answer: yes. Useless signal.
The real question is whether they can ship better, faster code with AI help. That needs a task, not a chat.
What to look for
A good AI-user does three things you can watch happen:
- Breaks a fuzzy task into small, verifiable steps before asking AI for code.
- Reads the AI output, finds the wrong bit, and pushes back.
- Tests what the AI gave them. They never paste and pray.
A weak AI-user pastes a vague prompt, accepts the first answer, and ships.
A simple test
Give the candidate a small broken project. Tell them they can use any AI tool they want. Give them 45 minutes to fix it and add one feature.
Then watch the recording.
Did they describe the bug to the AI before asking for a fix? Did they reject any answers? Did they run the code before committing?
Why EasyEnv fits this
EasyEnv records the full session: prompts, code, terminal commands, tests. You can replay how they thought, not just what they shipped.
A senior reviews the replay in ten minutes. AI fluency is no longer a vibe check.
When was the last time you watched someone use AI on real code, end to end?