AI Is Changing Work. What Are We Hiring For?

Define the contribution before specifying the tools

A requisition makes several operating choices, even when they remain implicit. It states what expertise the organization expects to acquire immediately, what it is prepared to develop, and what responsibility one person will carry. Adding AI terminology does little to clarify those choices.

The starting point is the result the role must help produce. What work will the person perform directly? What will they ask AI or existing automation to do? Which outputs must they verify? Where will they need other people, and which decisions remain outside their authority?

PwC’s 2026 Global AI Jobs Barometer, released on June 15, provides evidence that requirements are changing. In its analysis of 2.4 million US entry-level job advertisements, the most AI-exposed junior roles were seven times more likely than the least exposed to request traditionally senior skills. These advertisements reveal employer demand; they do not establish that applicants possess those capabilities or that interviews assess them effectively. [1]

The useful response is to make requirements more precise. “Strategic thinking” needs an observable meaning, just as a named platform requirement needs a reason. For a service-improvement role, that might mean recognizing that faster first replies will not improve resolution if requests still arrive at the receiving team without usable context.

The Outcome owner should explain the result and acceptance criteria. The hiring manager, working with HR and the relevant specialists, can then connect the role’s responsibilities to evidence a candidate could reasonably provide. That conversation belongs before the requirements become screening rules.

Separate essential entry capabilities from a workable learning commitment

An expert chef can use an unfamiliar recipe because experience helps them interpret it. They can notice when the ingredients, timing, or technique do not suit the intended result. Instructions become more useful when the person following them can recognize what needs adjustment.

My software-development foundation gives me a similar starting point when examining unfamiliar technology. I can investigate why a framework such as LangGraph exists, which problem it addresses, what alternatives might fit, and what would have to be demonstrated before relying on it. I can also ask whether the proposed work needs that kind of framework at all.

Those questions begin the investigation. Documentation, a bounded implementation, failure testing, and review by people with relevant experience help establish whether the understanding holds. A convincing specification is an intermediate result.

Research illustrates both the opportunity and its boundary. A field experiment published in Organization Science on June 12, 2026 found that AI-assisted individuals matched unassisted pairs on product-innovation challenges. Their proposals also combined technical and commercial perspectives more evenly. The participants were professionals working on problems within their organization; the result supports expanded reach in that setting, rather than unrestricted competence in a new profession. [2]

For each important requirement, the hiring team should decide what must already be dependable and what can be learned within the available time. A leader selecting an approach may bring relevant architectural and operating experience without prior use of the preferred library. An engineer expected to diagnose a production failure immediately may need substantial experience with the actual platform.

A learning commitment must be specific enough to rely on: who will coach the person, which work remains supervised, what practice is available, and what evidence will permit greater responsibility. Describing everything as learnable while providing no learning capacity simply transfers the unresolved risk to the new hire.

Make the assessment reveal understanding as well as output

Once AI can help produce a polished answer, the answer alone becomes less informative about what a candidate understands. Employers need evidence of how the person checks it, responds to uncertainty, and changes course.

A small randomized study published by Anthropic on January 29, 2026 examined 52 developers learning an unfamiliar Python library. The AI-assisted group scored lower on an immediate comprehension assessment on average, without a statistically significant time saving. Some uses of AI that involved active comprehension were associated with stronger learning, but those associations were exploratory. The study does not establish long-term learning effects. It does give leaders a reason to examine understanding separately from completed output. [3]

Established selection practice provides a foundation. The US Office of Personnel Management connects assessment to analysis of the job’s tasks and competencies. Its guidance on work samples emphasizes competencies expected on entry and cautions that the method may be unsuitable for activities that will be taught after selection. [4] [5]

Apply that distinction deliberately. If the job involves AI-assisted analysis, a relevant exercise can allow approved assistance and documentation. If independent diagnosis is essential during an outage, assess that capability under appropriate conditions too. Explain what each part is intended to establish.

Provide comparable tools, context, time, and familiarization. Define the evidence standard before observing candidates, and use assessors qualified to judge the work. OPM’s structured-interview guidance similarly uses predetermined questions and common rating standards. [6]

Ask the candidate to explain a consequential choice, show the source behind a claim, or revise the recommendation when a relevant condition changes. Look for justified acceptance of useful AI output as well as justified challenge. Eloquence, confidence, and the number of questions asked are weak substitutes for evidence that the person can carry the responsibility.

Follow one hiring decision through to the work

Hypothetical hiring example

Consider an organization hiring someone to improve its internal service-request operation. Employees receive quick acknowledgments, but resolution is unreliable. Requests move between teams, information is repeated, and exceptions wait for someone with the authority to decide.

The hiring manager gives candidates a bounded set of fictional cases, operating priorities, current guidance, and access to the same approved AI tool. The task is to recommend a useful first intervention and explain how its value would be tested. The exercise assesses capabilities identified as necessary on entry; unfamiliar product syntax is not itself a pass-or-fail condition.

A capable response might identify incomplete handoffs as the immediate constraint. The candidate could propose a small change to required information, supported by existing workflow rules, with AI helping prepare a summary where appropriate. A sound recommendation could also conclude that a simpler process change deserves testing first.

The assessor then introduces a new condition: the receiving team cannot provide review capacity at the proposed volume. Does the candidate revise the design, account for the additional workload, or narrow the trial? Can they explain what should happen to a request while it waits and who may authorize the next action?

The evidence is visible in the candidate’s choices. They connect the intervention to accepted resolution, identify the knowledge the work depends on, preserve consequential decisions with the authorized person, and make the receiving team’s capacity part of the proposal. They specify measures such as elapsed resolution time, incomplete handoffs, repeat contact, and total review effort. These are proposed tests, not achieved benefits.

This exercise would provide one part of the hiring evidence. Previous work, structured discussion, and appropriate references can help establish whether the demonstrated reasoning is consistent with experience. Its usefulness would need to be checked against the actual role; the example is not a validated selection instrument.

There is also an employer test. If the organization hires for this contribution, will the person receive the access, authority, and support needed to make it? Recruiting someone to connect an operation while measuring only the volume of locally completed tasks recreates the tension explored in AI Adoption: When the Old Way Still Wins.

Keep developing the expertise the operation will need

The strongest objection to changing technical requirements deserves attention. Specific experience often contains hard-earned knowledge of failures, dependencies, and recovery. A person who knows what breaks may recognize a problem that an impressive general explanation misses.

That is why the hiring decision should preserve depth where the work requires it. For each essential requirement, explain the failure or responsibility it addresses. For each requirement that can be relaxed, explain how capability will develop and what protection remains while it does.

The same care applies to broader traits. “Mindset,” “adaptability,” and “systems thinking” need observable expressions. A candidate demonstrates adaptation by revising a plan after investigating new evidence. They demonstrate systems thinking by identifying a dependency that changes the proposed action. These behaviors give assessment more substance than an interviewer’s impression of fit.

In an October 9, 2026 article published by the World Economic Forum, Ahmed Al Shamsi and Najwa Aaraj argue that problem definition and judgment become more important as AI improves. Their perspective reinforces a leadership question: how will the organization develop the experience on which that judgment depends? [7]

Hiring and development belong in the same operating decision. Experienced reviewers need time to coach. New colleagues need opportunities to investigate failures and explain their reasoning. Work must leave room to deepen the understanding required to supervise the assistance being used. Redefining Skill: Preparing People to Deliver with AI develops that learning responsibility; the AI-Era Workforce Capability framework connects it to the role and its operating conditions.

AI creates room to reconsider who could perform a role and how they might develop into it. Leadership must make the resulting expectations specific enough for both the candidate and the organization to meet.