The CHRO guide to AI

Your AI problem is not a technology problem and it will not be solved by a tools budget. It is a capability design problem, and it belongs to you rather than to the CIO, because every consequence that matters arrives in your function: who learns, who becomes senior, who is accountable for a decision a machine now shapes, and whether the organisation can still do the things it has quietly stopped practising. Three things have genuinely changed for HR since 2023. Adoption is effectively universal while the measurable impact on pay and hours is still close to zero, which means the window for design is open and will not stay open. The route by which juniors became seniors has been disrupted faster than any organisation has rebuilt it. And in the European Union, AI literacy became a legal obligation rather than a development ambition. What follows is what I would put on a CHRO's agenda, in the order I would put it.

What has actually changed

Adoption arrived faster than the technologies HR usually plans around. Bick, Blandin and Deming found that by late 2024 nearly forty percent of the US population aged 18 to 64 used generative AI, twenty-three percent of employed respondents had used it for work in the previous week, and nine percent used it every working day, with work adoption as fast as the personal computer and overall adoption faster than the internet. The same work found only one to five percent of total work hours being assisted, and reported time savings of about 1.4 percent. Enormous reach, thin penetration. Your people are using it. Your work has not been redesigned around it.

The outcome data says the same thing from the other side. Humlum and Vestergaard, linking adoption surveys to administrative records across roughly 25,000 workers in 7,000 Danish workplaces, found precise null effects on earnings and hours two years after ChatGPT launched, ruling out effects larger than two percent, while documenting substantial task reorganisation and entirely new tasks in content generation, AI oversight and AI integration. For a CHRO this is the most useful finding available, because it says the job architecture is moving before the compensation data moves, and job architecture is your responsibility.

The development question is where the evidence is sharpest and least comfortable. Brynjolfsson, Li and Raymond found AI assistance raised productivity by thirty-four percent for the newest staff and barely at all for the most experienced, which compresses the visible gap between a novice and an expert and makes output a much weaker signal of capability. And Bastani and colleagues, in a field experiment published in PNAS in 2025, gave nearly a thousand students access to a GPT-4 tutor: performance rose sharply while the tool was available, but students with unrestricted access scored seventeen percent lower than a control group once it was taken away, while a version designed to give hints rather than answers largely removed the harm. Translated into HR language: the design of the tool, not the presence of the tool, determined whether people developed.

On the design of oversight, Vaccaro, Almaatouq and Malone's 2024 meta-analysis of 106 studies and 370 effect sizes found human-AI combinations performing significantly worse on average than the better of human or AI alone, with losses concentrated in decision-making. Any policy in your handbook that says a human will review AI output is, on this evidence, not yet a control.

And employers, asked directly in the World Economic Forum's 2025 Future of Jobs report, name analytical thinking as the most valued core skill and skills gaps as the largest single barrier to transformation over five years.

The legal floor, and a change most guidance has missed

Article 4 of the EU AI Act, the AI literacy obligation, entered into application on 2 February 2025, and it reaches providers and deployers of AI systems, which includes most large employers using AI in their own operations. It has since been amended by the Digital Omnibus on AI, in force from mid-2026, and the change matters for anyone drafting policy. The current wording requires organisations to take measures to support the development of AI literacy among staff and others operating AI systems on their behalf, taking account of technical knowledge, experience, education and training and the context of deployment. The earlier text, which required ensuring a sufficient level of AI literacy, is what most published guidance still quotes. If your policy or your training vendor's materials cite the older phrasing, they are describing a superseded obligation. Supervision by national market surveillance authorities applies from August 2026.

Two cautions. First, this is the floor rather than the standard: a compliance-shaped training module satisfies a regulator and will not touch any of the capability problems below. Second, I am describing the obligation as the European Commission currently states it, not giving legal advice, and the position for your organisation should be confirmed with counsel.

The SuperSkills interpretation

The central shift I would ask a CHRO to make is from skills management to capability design. Skills management treats capability as an inventory: a taxonomy, a gap analysis, a catalogue of courses, a completion rate. That model was already straining before AI, and AI breaks it, because the thing at risk is not a missing skill that training can add. It is the erosion of judgement that was previously built by doing the work, and no course rebuilds it. Capability design asks a different question: given that the machine now does this, through what experience does a person still become capable of the senior version of this job? That is a work-design question wearing an HR badge, and it is the reason this belongs to you.

What accumulates when nobody asks it is what I call capability debt: the loss of human knowledge, skill and judgement that builds up when an organisation automates work faster than it redesigns how people learn through doing. It has three engines, and all three are HR processes. The missing rungs are the junior tasks that used to carry people upwards, removed by automation before anyone noticed they were load-bearing. The missed reps are the repetitions handed to the machine, so the work ships but the practice never happens. And synthetic seniority is the individual result: output that looks senior while the judgement underneath was never built, which your promotion process is currently unable to detect because it evaluates work product.

That last point deserves a blunt statement, because it is the one with the largest cost attached. Output quality has stopped being a reliable proxy for the capability of the person who submitted it, and virtually every performance and promotion system in existence rests on the assumption that it is. You are, right now, promoting on a signal that AI has degraded. Medicine and aviation solved this a long time ago by testing judgement directly, through live decisions, simulation and oral examination, rather than trusting that good work implies a capable person. Almost no corporate function does.

Six things to own

• The judgement map. A written statement of which decisions a human must own and why, produced with the business rather than for it. An organisation that has not written it has answered by default. See human and AI decision making.
• The development pathway. For each role where AI has taken the junior work, name what now builds the judgement that work used to build. If the answer is nothing, you have a leadership pipeline problem dated five to eight years out and no one else is tracking it.
• Assessment that survives AI. Move some evaluation from artefacts to judgement: live problems, reasoning made visible, the annotated draft showing what was prompted and what was changed. Otherwise you are assessing the model.
• Job architecture, deliberately. Apply the expertise test to each role: once AI takes the automatable tasks, is what remains harder or easier? Harder means fewer, better and better-paid, with a development plan. Easier means a role that is commoditising, with consequences for pay, standards and retention.
• AI literacy that goes past compliance. Meet Article 4, then teach the thing that actually protects the organisation: where the model is likely to be confidently wrong in your domain, and what to do about it.
• Measures that can fall. Adoption metrics only go up, which is why executives like them. Add at least one measure of capability without the tool and one of how often humans genuinely disagree with machine output. An approval rate near a hundred percent is a finding, not a success.

What not to do

Do not run a prompt-engineering training programme and call it a capability strategy. Tool fluency is being deliberately engineered to require less skill each quarter. It is worth an afternoon and it is not a plan.

Do not put a human in the loop and describe it as governance. Specify who, at what point in the process, with what authority to stop it, and measure the disagreement rate. The meta-analytic evidence says an undesigned pairing can be worse than either party alone.

Do not let the graduate intake be cut on a productivity argument without pricing the pipeline. The saving is this year and visible; the cost is the senior population in seven years and invisible. Somebody in the room has to hold the second number, and it will be you. See will AI replace entry-level jobs.

Do not wait for the productivity data. It is the slowest indicator available, and the Danish evidence is explicit that the structure of work moves well before earnings and hours do.

Development of the idea

The argument that HR should move from skills management to capability design is one I set out in WorldatWork's Workspan Daily in August 2026; that piece is member-facing, so it is described rather than linked here. In CEOWORLD in July 2026 I set out the drift versus design framework and the four postures organisations take. The accountability argument, including Human at the Start, human in the loop and human at the end, is in the European Business Review piece on accountability gaps in leadership decisions (21 August 2026). The full framework is in SuperSkills (Kogan Page, 2026). Rahim speaks regularly at HR and CHRO conferences on this material.

Key research and primary sources

• Bick, A., Blandin, A. and Deming, D. J. (2024). The Rapid Adoption of Generative AI. NBER Working Paper 32966.
• Humlum, A. and Vestergaard, E. (2025). Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI. NBER Working Paper 33777, revised March 2026.
• Bastani, H. et al. (2025). Generative AI Without Guardrails Can Harm Learning. Proceedings of the National Academy of Sciences, 122(26).
• Vaccaro, M., Almaatouq, A. and Malone, T. (2024). When combinations of humans and AI are useful: a systematic review and meta-analysis. Nature Human Behaviour, 8, 2293-2303.
• Brynjolfsson, E., Li, D. and Raymond, L. (2023). Generative AI at Work. NBER Working Paper 31161; published in the Quarterly Journal of Economics, 2025.
• European Commission. AI literacy: questions and answers, and AI talent, skills and literacy. On Article 4 of the AI Act.
• World Economic Forum (2025). The Future of Jobs Report 2025.

Related SuperSkills research

The organisation-wide version of this is AI workforce strategy, and the measurement argument is the AI readiness lie. On the capability mechanisms, capability debt, the missing rungs, synthetic seniority and how humans learn with AI. On oversight design, human and AI decision making and Human at the Start. The executive-team version is how leaders should respond to AI.

About this research

Rahim Hirji is the author of SuperSkills: The Seven Human Skills for the Age of AI (Kogan Page, 2026) and the founder of The SuperSkills Intelligence Company. This work draws on research across more than 200 organisations in 30 countries over seven years. Findings are attributed to the studies that produced them and kept separate from the interpretation, which is the author's. The description of Article 4 reflects the European Commission's current published position and is not legal advice. Capability debt, the missing rungs, the missed reps, synthetic seniority and drift versus design are part of the SuperSkills lexicon. This is a living reference, reviewed and updated as significant new evidence appears, and the regulatory section is on a 90-day review cycle.