How AI Changes Your Hiring Process — What to Hire For Now That the Old Job Descriptions Don't Apply
Founder, Prompt Consulting — AI implementation advisor for mid-market companies.
The competencies that defined a strong hire two years ago describe a smaller and smaller part of the work today. Hiring still works — but it works for a different shape of person. The change is more practical than philosophical.
A hiring manager at a marketing agency recently described her dilemma to me. She was interviewing for a senior content strategist role with the same competency matrix the agency had used for a decade. Strong portfolio, sharp writing samples, demonstrated craft. And yet she kept ending interviews unsure whether the candidates would thrive in the role as it actually existed now — a role where most first drafts came from an AI assistant, most of the day was spent editing and directing rather than writing, and the highest-leverage skill had quietly become "knowing what good looks like" rather than "producing it from scratch."
She wasn't alone. The hiring process at most companies hasn't caught up to how the work has changed. Job descriptions still list the activities people used to do. Interview rubrics still test for the skills that used to differentiate strong from weak candidates. And the candidates being selected are, increasingly, optimized for a job that doesn't quite exist anymore. The fix isn't dramatic. It's a series of small adjustments that, taken together, make the hiring process match the work again.
What the Work Now Requires
Before changing the hiring process, it helps to be honest about what the work has become. AI hasn't eliminated knowledge work; it has restructured it. The high-value activities have shifted, and the new shape is recognizable across many roles.
Judgment over production. Producing a first draft of anything — a memo, a model, a contract, a deck, a plan — is now mostly fast and cheap. The bottleneck has moved to deciding what to produce, evaluating whether the output is right, and editing it into something genuinely useful. Strong performers spend less time generating and more time judging.
Direction over execution. Senior individual contributors increasingly work like junior managers: setting up an AI to do a task, defining what good looks like, reviewing what comes back, redirecting when it's off. The shift is subtle but significant — the work pattern is closer to coaching than to doing.
Synthesis over recall. When information is instantly retrievable and instantly summarizable, the value of remembering things or being able to look them up declines. The value of putting things together — drawing connections across domains, integrating conflicting evidence, building a coherent picture from messy inputs — goes up.
Skepticism over acceptance. AI tools are confident. Some of their outputs are wrong. The cost of accepting a wrong output uncritically has gone up, because more outputs are flowing through the work and there's less time to scrutinize each one. People who default to verifying are differentially valuable now.
What to Test For That You Probably Aren't
If those are the high-value behaviors, the interview process should be able to detect them. Most don't. The standard formats — resume screening, behavioral interviews, take-home assignments — surface the older shape of competence and miss the newer one. A few targeted additions help.
Editorial judgment exercises. Give the candidate a piece of AI-generated work in their domain — a draft strategy memo, a model analysis, a plan document — and ask them to assess it. What's strong? What's wrong? What's missing? What would they change before using it? This reveals judgment in a way that asking them to produce something from scratch does not.
Prompt-and-evaluate exercises. Ask the candidate to use an AI tool to accomplish a task in front of you, and watch how they work. Do they set up the task well? Do they read the output carefully? Do they iterate productively or accept the first answer? This is increasingly the actual work, and watching someone do it is more informative than asking about it.
Verification probes. Ask a question with a confident, plausible, wrong answer somewhere in the candidate's domain. See whether they push back, qualify, or accept. Candidates who default to accepting confident-sounding statements will do the same with AI outputs.
Synthesis tasks. Give the candidate multiple sources with overlapping and conflicting information and ask them to build a coherent view. The depth and structure of their synthesis tells you more about their thinking than any single-source answer would.
What to Stop Over-Weighting
The other side of the change is letting go of signals that mattered more when the work was different. Continuing to optimize for them filters out strong candidates without helping you find better ones.
Pure production speed. "Can produce a polished memo in two hours" was a useful proxy for skill when polishing took most of the time. Now AI can produce the same draft in two minutes, and the candidate's two-hour version isn't a better signal of capability — it might be a worse one if it suggests they're attached to a workflow that's been superseded.
Domain knowledge as recall. Knowing the relevant facts of your industry remains useful but is no longer a differentiator. The candidate who knows everything about your domain off the top of their head and the candidate who can pull and synthesize the right context in five minutes will perform very similarly in the actual job.
Polished writing samples in isolation. A perfect cover letter or writing sample now tells you less about the candidate's writing than about their AI usage. This is not necessarily bad — using AI well is itself a skill — but treating polished output as direct evidence of personal capability is a category error you can avoid with the assessment changes above.
Years of experience as a proxy. Experience still matters, but its value has compressed. Five years of experience accumulated under the old work patterns may transfer less directly than two years of experience accumulated in the new patterns. Calendar time is a noisier signal than it used to be.
New Roles That Have Emerged
Beyond reshaping existing roles, AI has created some new ones — or made previously informal roles into formal ones. Worth knowing what they are, because the same companies often try to fit these into legacy titles and get confused results.
AI workflow designers. People who design the prompts, tools, and process around AI use for a specific function. Not engineers — they don't write code. Not pure operators — they think in systems. The best are deeply familiar with one functional area (sales, finance, support) and have developed strong instincts about what AI can and can't do well.
AI editors and reviewers. People who specialize in catching the failure modes of AI output for a specific domain. Legal AI reviewers, medical AI reviewers, technical writing AI reviewers. The role is half quality control, half subject matter expertise. Increasingly important as AI output volume scales.
Automation product managers for internal use. People who own the productivity tooling for an internal function, treating it as a product with users, requirements, and a roadmap. This role used to be informal — somebody on the team who happened to be good at tools. It's increasingly formal because the leverage is large.
Data and knowledge stewards. People who own the quality and curation of the data and knowledge an AI system uses. Without them, AI quality drifts. With them, AI gets better over time. Often underestimated in headcount planning because the work is invisible until it isn't being done.
Practical Adjustments to Make Now
The implications for hiring aren't theoretical. There are specific changes that move the process in the right direction without throwing out what already works.
Rewrite job descriptions for current work, not legacy work. Walk through what people in the role actually do today, not what the JD has said for the last three cycles. The gap is usually large. Updating the JD changes who applies, and a different applicant pool is most of the battle.
Add at least one AI-relevant exercise to every interview loop. Whatever the role, there's a version of the work that involves AI. The interview should include it. Doing this reveals candidates who are comfortable with the actual work and surfaces those who would struggle once hired.
Train interviewers to evaluate AI-assisted output. Most interviewers were trained to evaluate unassisted output. Evaluating AI-assisted work — what the candidate added, what they caught, what they decided — is a different skill. Without training, interviewers default to discounting AI-assisted work or accepting it uncritically.
Reconsider the seniority ladder. The work that used to take a junior person two years to become competent at can now be done with much less experience using AI well. The work that used to take a senior person one hour now takes ten minutes, but the judgment to do it well still takes experience. The ratio of junior-to-senior headcount that the work supports has changed, and your hiring plan should reflect that.
Stop hiring for the role the company had three years ago. This sounds obvious and is consistently violated. Replacement hires often default to "find someone like the person who left." But the role has often changed since that person was hired. The replacement should fit the role as it is, not the role as it was.
What Changes If You Get This Right
The companies that adjust their hiring well over the next eighteen months will end up with a workforce that's better fitted to current work. Their interviewers will be assessing the things that matter. Their job descriptions will attract candidates suited to the actual job. Their seniority structures will match the leverage the new work patterns enable. None of this is dramatic in any single hire; it compounds across a hundred hires into a meaningfully different organization.
The companies that don't adjust will find themselves with talented people in the wrong configuration. Good candidates filtered out because they didn't match an obsolete profile. Hired candidates struggling with work the JD didn't describe. Senior people doing work the role no longer requires while AI-leveraged junior work goes underutilized. The cost shows up as turnover, frustration, and a vague sense that hiring is harder than it used to be.
The work has changed. The hiring process should change with it. The candidates are out there — the question is whether your process is set up to recognize them.