Learning to use AI does not guarantee a say in how work changes. The next workplace bargain must protect opportunities to exercise judgment, acquire skill and make a contribution people recognise as their own.

On October 13, Elon University's Imagining the Digital Future Center published a survey with an uncomfortable pairing: 66% of American workers were comfortable using AI at work or open to doing so with training, while 52% of all adults expected AI to leave most workers with little control over their futures. Conducted in August, the survey measures expectations, not a demonstrated collapse of employment or purpose. But it challenges a comforting assumption: willingness to learn a technology means accepting the working life built around it.

My argument is that a successful AI transition must preserve meaningful discretion inside jobs that survive. Training matters. So does having something consequential left to decide.

Consider two possible futures for a claims handler. In one, AI assembles documents and flags inconsistencies, leaving her more time to understand an unusual case and explain a decision. In another, software supplies the answer, measures how quickly she approves it and treats departures as poor performance. Both employers could report faster processing. Only the first design clearly expands her capacity to contribute. These are illustrative choices, not accounts of a particular employer.

The difference is not sentimental. Judgment develops through practice, correction and responsibility. If junior employees mainly approve machine suggestions, an employer may save money today while weakening the experience from which tomorrow's experts emerge. If employees' accumulated knowledge becomes embedded in software while their discretion shrinks, their bargaining position could weaken even without a redundancy notice.

There is evidence that speed and richer work do not automatically travel together. In findings published on October 6, Gallup reported that 63% of surveyed American employees who used AI said it helped them work faster; 47% said it let them spend more time on interesting work. These are self-reports from earlier this year, not causal estimates. Regular use was also more common among graduates and managers. Access to benefits may therefore follow advantages people already possess.

The serious counterargument is that much work is repetitive, frustrating or subject to arbitrary human supervision. Removing it can improve lives. An OECD survey of managers across six countries found that 60% perceived better decision-making from algorithmic management. Such systems include ordinary software as well as AI. The same research records concerns about unclear responsibility, hard-to-understand decisions and workers' health. Neither managerial enthusiasm nor anxiety settles the outcome for employees.

Employers should therefore examine who gains authority when they buy efficiency. Better scheduling could reduce wasted journeys and chaotic shifts. But a system can also make one supervisor responsible for more people while giving each worker less room to explain an exception. Licence fees are only part of the cost: training, integration and time for human review belong in the business case. Otherwise, apparent savings may depend on employees absorbing mistakes they cannot correct.

India's relevance extends beyond office workers using chatbots. ILO–JRC case studies examined logistics and healthcare in India, South Africa, France and Italy, finding efficiency benefits alongside risks to job quality and intrusive surveillance. This is qualitative evidence about management technology, not a national measure of generative AI's effects. It nevertheless shows why the debate must include people managed by software as well as people who operate it.

For an Indian employer, a practical test is whether a worker can correct a misleading score without losing a day's pay in the process. Explanations should be available in a language the worker understands, with an accessible human route when connectivity or disability makes the interface difficult. These are design requirements worth testing, not benefits already established by the research.

Three futures remain plausible. In the better one, AI removes drudgery while workers retain responsibility, learn and advance. In the uneven one, senior employees gain capable assistants while others become approval clerks. In the harmful one, opaque targets intensify work and make dissent expensive. The dividing line is partly what employers reward: sound judgment and development, or compliance at speed.

After deployment, which decisions can frontline workers still make? What will a new recruit learn to do independently? Can employees challenge the system, and does anyone have paid time to listen?

Ask for answers before approving the productivity forecast. Employers should agree retained responsibilities with workers and review learning, discretion and promotion opportunities alongside output. Policymakers should examine the quality of surviving jobs as well as how many disappear. A job worth keeping should leave its holder with a recognisable part in doing it well.