You already know which jobs AI will change. Do you know who's ready?

I've been reviewing a lot of versions of the same exhibit: a heat map of jobs or job families, shaded in different colors based on how exposed each one is to AI. There are a variety of methods used to predict what the new organizational structures will look like. While that's important data, here at Vervoe, we've also been focused on understanding readiness for people in those roles to do the new version of the work.

Many organizations believe they know which jobs will change with AI, and they may be broadly right. However, a forecast about a role is not an assessment of a person. To know who's ready, you have to define what "ready" means in the changed version of the work, then observe people doing that work and score it against a consistent standard.

Exposure is a forecast. Readiness is a measurement.

We find it useful to separate two kinds of claims that often get blended together.

The first is a claim about roles and tasks. The World Economic Forum's Future of Jobs Report 2025, based on a survey of more than 1,000 employers, found that employers expect 39% of workers' core skills to change by 2030. Task-level research points the same way: Eloundou and colleagues estimated that around 80% of US workers could have at least 10% of their tasks affected by large language models.

Neither of these studies tells us anything about a specific claims analyst, account manager, or engineer in your organization.

AI readiness is the demonstrated ability of a specific person to perform the AI-changed version of their role to the required standard, including the judgment to recognize when AI output is wrong.

Readiness is role-specific, so there's no universal readiness score. It's demonstrated, so it has to be observed rather than claimed. And it includes judgment, so knowing how to use a tool is necessary but not sufficient.

Why surveys, usage data, and training completions aren't readiness evidence

When organizations try to gauge readiness, they usually reach for data they already have. Each source tells you something real. None of them tells you whether a person can do the changed work well.

• Self-assessment surveys
  What it can tell you: Confidence, attitudes, and perceived development needs
  What it can't tell you: Whether the person can do the work to standard
• Usage analytics
  What it can tell you: How often tools are used, and for which tasks
  What it can't tell you: Whether the output was good, or when AI should have been overridden
• Training completion
  What it can tell you: Exposure to learning content
  What it can't tell you: Whether learning transferred to the job
• AI knowledge tests
  What it can tell you: AI literacy: concepts, terms, and basic practices
  What it can't tell you: Applied performance in a specific role
• Task-based assessments
  What it can tell you: How a person performs realistic, AI-integrated work against a defined standard
  What it can't tell you: Future potential with certainty, or context the task didn't capture

Consider the most-cited number in this space. The EY 2025 Work Reimagined Survey of 15,000 employees and 1,500 employers across 29 countries found that 88% of employees use AI at work, mostly for basic tasks like search and summarization. Only about 5% qualified as advanced users. But since this data is self-reported, like most evidence in this space, our best picture of the readiness gap rests on people accurately describing their own behavior.

Usage data has a different problem: it measures frequency, not quality. And frequency can mislead. In a field experiment with 758 BCG consultants, Dell'Acqua and colleagues (2023) found that consultants using GPT-4 did substantially better on tasks inside the AI's capabilities. On a task outside them, they were 19 percentage points less likely to reach a correct answer than those working without AI. This data highlights an important risk.

The construct problem: ready for what, exactly?

You can't measure a construct you haven't defined. For AI readiness, that means you need a clear picture of the target role before you assess anyone against it.

That's harder than it sounds. In traditional selection, we build criteria from the job as it's performed today, drawing on incumbents, performance data, and subject-matter experts. For an AI-changed role, parts of the target job don't exist yet. There may be no incumbents doing it at scale and no performance history to learn from.

Industrial-organizational psychology has tools for this, and so does Vervoe's Skills Advisory team. Future-oriented job analysis brings together subject-matter experts, early adopters already working with AI, and the leaders accountable for outcomes. Together they specify the changed work: which tasks shift, where AI enters the workflow, which decisions stay with people, and what "good" looks like. Without that step, a readiness assessment is measuring something, but no one can say what.

A practical framework: four questions to answer before you measure anyone

1. What will the changed work actually require? Document the tasks, AI touchpoints, decision rights, and quality standard for each priority role. Revisit it as the tools change.
2. What evidence would convince a skeptical expert? Design tasks that mirror real AI-integrated work, such as correcting an AI-drafted client response or deciding when a model's recommendation shouldn't be followed. Score them with rubrics anchored in observable behavior.
3. Does the measure hold up? Check consistency, gather evidence that links each task to the job analysis, and monitor subgroup differences. Provide accommodations, and tell employees what's being measured and why.
4. What decision will the results inform, and who owns it? Using results to target development is lower stakes than using them for redeployment or role changes. The higher the stakes, the stronger the evidence and governance you need, and the more human review matters.

Why task-based evidence is the scientist's default

When the question is "Can this person do this work?", the most direct evidence comes from watching them do a representative sample of it. That's the logic behind work samples and job simulations, and the research supports it. In their 2022 re-analysis of selection research, Sackett, Zhang, Berry, and Lievens estimated an operational validity of about .33 for work sample tests, which places them among the more predictive selection methods.

Where Vervoe fits

Vervoe is a skills validation platform built on a simple premise: people should be able to prove what they can do. The same approach that lets organizations evaluate candidates through realistic, job-related tasks extends to the existing workforce, including AI readiness assessments that place employees in AI-integrated work scenarios.

Vervoe scores responses consistently at scale against structured rubrics tied to job-relevant skills, and links each score to the response evidence that produced it. That turns demonstrated performance into decision-ready intelligence about where AI capability sits and where the gaps are, while decisions stay with the people accountable for them.

For organizations still defining what their AI-changed roles require, Vervoe's Skills Advisory team works with talent and transformation leaders on the harder upstream questions: defining the skills that matter, designing measures that are fair, and building processes that hold up to scrutiny.

What to do in the next quarter

1. Choose two or three job families where AI exposure and business impact are both high.
2. Run a future-oriented job analysis with experts and early adopters to define the changed work and its standard.
3. Build or adapt task-based assessments for those roles, then pilot them with a representative group.
4. Review the pilot data: reliability, score distributions, subgroup differences, and employee feedback.
5. Use the first round to target development, not to make consequential decisions, until the evidence supports more.
6. Re-measure on a set cadence, because the tools and the work will keep changing.

A forecast tells you where to look. Measurement tells you what you'll find.

Knowing which jobs AI will change is a necessary first step, and many organizations have taken it. The next step turns strategy into action: finding out, role by role and person by person, who can already do the changed work and who needs support to get there. That means defining readiness carefully, measuring it through realistic work, and governing the results responsibly. It's harder than commissioning a heat map. It's also how you move from assumption to evidence.

See how Vervoe's AI readiness assessments turn this framework into a working pilot for your priority roles.

Sources

• Zell & Krizan, "Do People Have Insight Into Their Abilities? A Metasynthesis," Perspectives on Psychological Science (2014): across 22 meta-analyses, the mean correlation between ability self-evaluations and performance was .29.