When the Interview Becomes the Test: Building a Body of Work in AI Direction

The résumé has long been a curated artifact. Increasingly, it is being supplemented by something far less forgiving: a live demonstration. Across a growing number of US employers—from Chicago marketing agencies to Austin-based fintech startups—candidates are being handed a laptop, a scenario, and thirty minutes with a generative AI tool. The prompt is the interview now.

For years, job seekers could list 'AI proficiency' as a bullet point and move on. That era is closing. When a skill becomes cheap to claim, employers stop trusting claims and start testing behavior. The question is no longer whether you use AI, but whether you can direct it competently while someone watches.

What the Live Test Actually Measures

Hiring managers we speak with are surprisingly consistent about what they're evaluating, and it is rarely the cleverness of the prompt itself. They are watching for three things.

First, problem decomposition. Strong candidates break a vague request ('draft a customer retention email') into the components the model actually needs: audience, tone constraints, regulatory language, the specific action the customer should take. Weak candidates type the vague request verbatim and hope.

Second, iterative refinement. The one-shot prompt is a lottery ticket. Skilled practitioners treat the first output as a draft, interrogate it, and steer a second pass. Observers can tell within ninety seconds whether a candidate is collaborating with the tool or merely waiting for it.

Third, judgment about output. The differentiating candidate doesn't just produce text—they critique it. They catch the hallucinated statistic, flag the tone mismatch, and note what would need human verification before anything ships. That final instinct, more than any prompt technique, is what separates fluency from familiarity.

The Case for a Prompt Portfolio

If live tests are coming, professionals should not wait to be tested cold. We recommend assembling what might be called a prompt portfolio: a private, curated record of your best AI-directed work.

A credible portfolio contains three layers. The first is before-and-after artifacts—a mediocre first draft and the refined final version, with the prompts that moved between them. The second is constraint documentation: notes on how you handled a client's compliance language, a brand voice guide, or a tight word limit. The third is failure cases. Counterintuitively, showing where an AI approach failed and how you detected and corrected it signals more maturity than ten clean successes.

This portfolio serves a dual purpose. In interviews where no live test occurs, it becomes evidence. In interviews where one does, the practice of building it is itself the rehearsal.

Practicing Under Pressure

Prompting in a quiet office and prompting in front of a hiring panel are different sports. Pressure degrades the exact habits that matter—decomposition, iteration, verification. Three exercises build tolerance.

The twenty-minute drill. Take a realistic task (a project kickoff memo, a competitor summary), set a timer, and complete it end-to-end with an AI tool. Repeat weekly with new scenarios. Speed without sloppiness is trainable.

The constraint ladder. Redo the same task under progressively tighter limits: fewer turns with the model, no follow-up questions, a mandated output format. Constraints force the upfront thinking that strong prompting requires.

The red-team review. Have a colleague take your final output and attempt to find errors. Getting comfortable having your AI-assisted work scrutinized is precisely the comfort a live test demands.

Beyond the Buzzwords

Employers are not looking for prompt poets. They are looking for professionals who treat AI as a capable but fallible collaborator—people who give it context, push back on its output, and own the result. That is not a technical skill. It is a communication discipline, and like all communication disciplines, it improves with deliberate practice and honest evidence.

The professionals who build that evidence now—before the laptop is slid across the table—will find that the live test is not a threat. It is an opportunity to demonstrate something their peers can only claim.