AI agents at work

Best as partners. Studies of AI help at work find real gains on tasks within the AI's range, largest for less experienced workers: support staff solved 15 per cent more problems per hour and writing tasks took 40 per cent less time. Working together beats people alone but often not the better of the two, and does best on creative work. Workers want agents to take repetitive tasks and leave them as equal partners.

What the report found

1. Agents are moving from trial to daily work. In a survey of 31,000 workers in 31 markets, eight in ten leaders expected agents to be part of their company's AI plans within a year or so. The share of large organisations scaling agents rose from about a quarter to four in ten in a year.
2. Older research on automation still applies. Decades of studies show that trust works best when it matches what a system can do, that people who only watch a system lose track of it, and that people value keeping some control, even a little.
3. Early studies find real gains, largest for newer workers. In one study of more than 5,000 support staff, AI help raised the number of problems solved per hour by 15 per cent on average, with the largest gains for the least experienced. In a writing experiment, tasks took 40 per cent less time.
4. Working together is not automatically better. A review of 106 experiments found that people and AI together beat people alone, but often not the better of the two working alone. Combinations did best when people were stronger at the task and when the work was creative.
5. Workers want a partner, not a replacement. Across 104 occupations, workers welcomed automation for nearly half of their tasks, mostly to free time for more valuable work. Equal partnership was the most wanted arrangement, and clear instructions keep delegation honest.

What is inside

1. What is new
2. What older research tells us
3. What early studies suggest
4. What nobody knows yet
5. Questions worth watching

Plus an evidence maturity map, the Human Agency Scale, a scenario grid, a list of signals to watch and five tests for an AI agent that serves the people who use it.

Who it is for

For managers, team leads, designers of AI products, HR and learning teams, and anyone handing work to software that acts for them.

How we work

Frontier Reports separate what is established from what is emerging and what remains untested. Every claim is graded, and where researchers disagree, as they do on whether people and AI agents do better together, both sides are set out before we say where we stand.

Cite this report

The Why We Project (2026) AI agents at work: Software that acts for us. Frontier Report No. 21. Dubai: Why We Tech Ltd. Available at: thewhyweproject.com