Talent Screen

Screen candidates or current team members for AI leverage. Start with disposition: the temperament to delegate, iterate, and treat the tool as a collaborator. Then look for workflow design: the ability to turn that work into a recurring job with a check and an owner.

What you’re screening for

The thing that separates high-leverage AI users from everyone else isn’t tool knowledge. It’s disposition: the temperament to delegate work to a model, iterate on imperfect outputs instead of bouncing, and treat the tool as a collaborator rather than a search engine.

That gets someone through a strong session. The next level is workflow design: recurring work that reaches other people without the designer having to run it each time, with a check and an exception path that make the result safe to use.

Tool fluency is a week of practice. Disposition doesn’t come from training. You’re looking for people who already think this way, not people you can teach to think this way.

Don’t be fooled by resumes listing “GenAI tools,” prompt engineering certifications, or AI bootcamp credentials. These correlate with awareness, not leverage. The person who lists five AI tools on their resume and the person who quietly does the work of two people using one tool are different people. You want the second one.

What you need from the user

Ask for three things:

1. Role. What job function? (Engineering, marketing, finance, legal, ops, sales, etc.)
2. Seniority. Junior, mid-level, senior, or director+. The questions stay the same but the rubric for good answers shifts.
3. Context. Are they hiring a new candidate or assessing someone already on the team? The delivery format changes.

The four questions

Tailor these to the role. The structure stays the same; the examples should reference work specific to their function.

Question 1: “What’s the last thing you built or changed in your workflow because of an AI tool?”

What good answers sound like: specific tool named, specific workflow described, clear before-and-after, slight irritation about the part that’s still not solved. The specificity is the signal. “I use ChatGPT for brainstorming” is not a good answer. “I rewired how I prep for quarterly reviews. I feed Claude the last quarter’s metrics and my team’s project notes and it drafts the narrative section. Cut prep from six hours to forty minutes, but I still rewrite the recommendations because it gets the political context wrong” is a good answer.

For junior candidates, lower the bar on scope but not on specificity. A junior who automated their own onboarding notes is showing the same disposition as a senior who restructured a department workflow.

Question 2: “Tell me about a time an AI tool gave you a wrong answer. What did you do?”

What good answers sound like: they iterated. They didn’t bounce. They describe refining the prompt, providing more context, trying a different approach, or using the wrong answer as a starting point they edited. The key signal is the wrong answer didn’t make them stop using the tool.

Red flag: “It kept giving me bad answers so I stopped using it.” That’s someone who expects tools to work on first try. They won’t develop leverage.

For senior candidates, also listen for whether they built something to prevent the error class from recurring (a prompt template, a verification step, a workflow change).

Question 3: “If I gave you budget for any AI tools and some time to redesign your work, what would you spend on first? What would you stop doing?”

What good answers sound like: a specific opinion. Not “it depends on the strategy” or “I’d need to compare the options.” They name a tool or a workflow. They have a take on what’s wasteful in their current process. They’ve already thought about this.

For director+ candidates, listen for whether their answer extends beyond their own work to their team’s. “I’d give my three strongest people API access and have them build templates for the rest of the team” shows systems thinking.

Question 4: “Tell me about a piece of recurring work you got a tool to do on its own. Who benefits from it now, just you or others?”

Then ask: “How did you know it was right before people relied on it? What can two people check and agree on? What happens to an exception?”

What good answers sound like: a real job on a cadence, a clear trigger and audience, a mechanical check, and a named exception owner. The candidate who automated their own inbox is useful. The candidate who set up a job that produces a checked report for a team, without needing to be present, designs workflows.

For junior candidates, a small recurring workflow can count. Scope is not the test. Evidence that they thought through the check and exception path is.

Scoring rubric

Generate a rubric tailored to their role and seniority. Score session leverage and workflow design separately. Do not average them: a fast AI user is not necessarily someone who can design work that runs without them.

Session leverage

Strong signal (recommend hire / high-leverage team member):

• Answers Questions 1–3 with specific examples
• Describes iterating on wrong answers, not abandoning
• Has an unprompted opinion about next tool or workflow change
• For existing team members: teammates name them as the person to ask about AI

Moderate signal (proceed with exercise / developing leverage):

• Answers one or two questions with specificity
• Uses AI tools but hasn’t reorganized workflow around them
• Open to iteration but hasn’t built the habit yet

Weak signal (pass / not a champion candidate):

• Answers in generalities (“I use AI for productivity”)
• Cites a single bad experience as reason tools don’t work
• Talks about AI in future tense despite having had access for months
• Hasn’t opened the tool in the last week

Workflow design

Strong evidence:

• Names recurring work that runs on a cadence for a clear audience
• Explains the trigger, a mechanical check two people could agree on, and the exception owner
• Can show how the workflow keeps working when they are out of the loop

Partial evidence:

• Has automated personal work but cannot name the audience, check, or exception route
• Has a sensible idea for a recurring workflow but has not built it yet

No evidence:

• Describes a one-off AI session as automation
• Says a senior person would review every result instead of naming a check
• Cannot say where a failed check goes

Unattended work that others will depend on needs a brief governance follow-up: a named owner, access that survives the designer, a visible check or log, and an exception and shutdown path. Record that follow-up in the screen. Do not turn the screen into a policy audit.

The structured exercise (hiring only)

If the user is hiring, suggest a paired exercise: give the candidate a real, non-confidential task from the role. 90 minutes. Any tool allowed. Watch the workflow, not just the output.

What to observe:

• Did they pick a tool with intent, or fumble around?
• Did they iterate when the first result was wrong?
• Did the output improve over the session?
• Could an unaided person produce the same result? (If yes, AI didn’t create leverage.)

At minute 75, give them a second prompt: “Name one part of this task you would put on a cadence. In five lines, name its trigger, audience, the mechanical check that shows it is right, and the exception owner.” Score this response separately from the session output. “I would review it” is not a check. A workflow designer can name the test and where a failed test goes.

Tailor the exercise description to the role. For engineering: a debugging or refactoring task. For marketing: draft campaign copy from a brief. For finance: analyze a dataset and produce a summary. For ops: design a process improvement from a problem description.

What to output

For hiring: A document with the four tailored questions, separate session-leverage and workflow-design scores calibrated to seniority, and a suggested exercise with criteria for both the 90-minute session and the minute-75 design prompt. When a candidate describes unattended work others may depend on, include the short governance follow-up. Title it “AI Leverage Screen: [Role Title].”

For team assessment: A document with the four questions reframed for a 1:1 conversation (less formal, more exploratory), both scores, and a recommendation section:

• People strong on both signals are your champions. Give them explicit air cover, a small tool budget, and a public expectation that the team adopts their artifacts.
• People strong in a session but without workflow-design evidence can do valuable work. Do not assume they can build unattended workflows for the team yet.
• People who score “moderate” will follow the champions if the signal is right. Don’t spend energy converting them directly.
• People who score “weak” after six months of access won’t develop leverage through training. Staff them on work where existing experience compounds. Don’t put them on operating model redesign.

Don’t sugarcoat the weak-signal assessment. A meaningful fraction of any team won’t develop AI fluency. That’s a staffing input.