AI Skill Reviews Force a New Arithmetic on Jobs
Flood tide: a push with no paper trail
No memo at the large consultancy tells Pamela to use artificial intelligence. Pamela, a senior executive in the United States, says there is no formal mandate ordering employees to learn AI. The ground is shifting under her anyway. In performance reviews, the people who show fluency and fluidity with AI are the ones who receive the rewards. It is quiet, she says; nobody is saying "learn AI or else." Yet at the annual review, managers reward whoever uses the tools well. The same job title and the same tenure now carry a different value. Colleagues who treat AI as a threat and those who treat it as a tool receive different attention. Once leadership notices the gap, Pamela says, it widens quickly. Employees who are not visibly using AI face slower promotion timelines. It becomes harder to be seen and harder to fight being on a shortlist. Some companies have abandoned the quiet approach altogether. Julie Sweet, the chief executive of Accenture, said in March on the Rapid Response podcast that AI at Accenture is now how work is done. [1] To be promoted, a consultant has to operate in the way Accenture now operates, meaning through AI. Disney, Meta, JP Morgan and KPMG have built "AI leaderboards" that track and rank employee use of the available chatbots and large language models. [2] The rankings show who uses the tools most, or at least most visibly. Coinbase turned the expectation into a dismissal trigger. Its chief executive, Brian Armstrong, requested that engineers complete AI training. Engineers who did not finish the training lost their jobs. The two approaches -- one silent, one announced -- add up to the same rule: the machine has entered the employee's job description. In Spain, Duncan Trevithick, 34, works in marketing for a company that provides AI training data. His employer has invented a different incentive. If Trevithick uses AI to complete more tasks faster, he qualifies for a bonus at the end of the year. On the surface, the deal is generous: the employee is paid to become better at using the most important tool of the day. Trevithick is not convinced. The uncomfortable interpretation, he says, is that employees are being assessed on how effectively they can take part in their own redundancy. He calculates what would happen if AI freed up two days of work each week. He would not receive two days off. He would not receive a 40% pay rise. The higher output would simply become the new baseline for what his employer expects. In the short term, that might help him get promoted. In the longer term, he has helped prove how much of his job no longer needs him. To steady himself, he tries to think like a manager again: he manages people, he manages AI agents, and he tries to own the loops between them.
Slack water: a return that has not arrived
Why would executives push AI so hard if they are not sure of the payoff? Consultancy McKinsey holds up the number that unsettles them. It reports that 94% of companies have yet to see significant value from AI. The firms pouring money into chatbots, training and leaderboards are, by their own accounting, still waiting. Only a small fraction of companies report value worth noting. That gap between investment and result is what makes employee usage attractive as a metric. Usage can be counted today, while business value remains a project for next quarter. Executives want to show progress on their AI budgets, and the performance review is the most direct scorecard in front of them. The pressure arrives at a moment when moving to a different employer is hard. In the United Kingdom, job vacancies have reached a five-year low. When jobs are scarce, workers cannot easily walk away from a review system that contains an AI criterion. The recruiter Gi Group polled 1,881 British jobseekers in July. [6] Of those, 75% said they would not be put off applying to an organisation that had added AI proficiency to its individual performance reviews. [6] Around 22% said the new criterion would put them off. [6] That leaves the majority treating AI proficiency as an entry fee. In a weak labour market, they are in no position to haggle over the currency. The legal boundary lags behind the culture. Tina Chander, a partner and employment lawyer at Weightmans, says employers are within the law when they rewrite review criteria. [7] Legally, they are free to define what a good employee looks like, and they may decide that AI skill belongs on that list. Chander's questions are about fairness once AI makes the employee faster. If the employer raises expectations because the employee can now produce more, is that fair? If the employee who is more efficient is rewarded differently, should a promotion suffice? Chander also sees a hidden disincentive. An employee who works faster with AI is, in effect, demonstrating that part of the job no longer needs a human hand. Why would any sensible worker volunteer for that demonstration? She advises employers to define policies, train their people and draw boundaries before the questions become disputes. If the policies stay vague, she says, employees will bring claims about fairness, performance expectations and job security. [7] Those claims will become easier to bring in the United Kingdom from January 2027. Under current rules, an employee needs two years of service before claiming unfair dismissal. The claim must be filed within three months of leaving. In January 2027, the required service drops to six months. The filing window doubles to six months. An employee dismissed after a negative AI performance review, perhaps only six months into the job, will have a longer runway to challenge the decision. Chander's advice about clear policies is not abstract; it is a preparation for that calendar.
Ebb tide: metrics in retreat
If employers wanted to defend their AI push, they would need something that is often missing: an honest explanation of its purpose. Tina Rahman, an HR consultant who runs the London-based consultancy HR Habitat, sees the absence every day. She says employers are not transparent enough about why they want AI inside people's jobs. The actual end purpose, she believes, is to save costs, save time and cut down on outsourcing. Companies rarely tell workers that. Partly they misunderstand their own tools, and partly they fear the reaction. The lack of transparency creates an information vacuum. Employees are left to decide whether the new review metric is a career chance or a career warning. Without a stated reason, the warning often wins. One danger of a universal AI mandate is that it turns work into theatre. Kamila Miller, applied AI researcher and lecturer at Henley Business School and based in Reading, has studied compliance up close. [8] Make AI usage a key performance indicator, she warns, and employees will log their interactions because they need the metric. They will route simple tasks through a chatbot that never needed one. They will generate AI-flavoured outputs that look productive on a dashboard. The company will measure adoption and mistake it for progress. Miller says the company will not measure judgement, learning or better decisions. Her verdict is direct: "You have not made people more skilled - you have made them more obedient." [8] [Pic1] Two prominent firms have already pulled back from the metric. In April, Luis von Ahn, the chief executive of Duolingo, told the Silicon Valley Girl podcast that AI use no longer counts in performance reviews at his language learning company. [9] Employees had started asking whether they were expected to use AI for AI's sake. Duolingo retreated and set a simpler rule: the most important thing is doing the job as well as possible, with or without AI. The Financial Times reported in May that Amazon shut down an internal leaderboard tracking employee AI use. The reason: employees were gaming the rankings by setting AI needless tasks. When the score is visibility, people perform for the dashboard instead of for the customer. Amazon and Duolingo each learned that the instrument meant to measure skill was measuring something else. So the people inside the system are drawing their own conclusions. Pamela, in her mid-50s, plans to retire within ten years and wants to keep the pension and health benefits tied to her job. She is visibly stepping up how AI helps her win new clients and makes sure she is credited for it. Trevithick, in marketing, is building side projects in the hours that AI saves him. His reasoning follows the capitalist logic he works inside. If AI can do a job better and cheaper, a business will replace the person doing the job; no appeal changes that arithmetic. So he aims to move towards a position where he owns assets, where he can leverage AI, and where the machine works for him. Can any performance review tell the difference between a worker replacing herself and a worker building something next to the machine?