Companies test AI agents at work, but most are still cautious

In short: More companies are trying “AI agents” at work, but most businesses still have not rolled them out widely.

What's going on

AI chat tools usually answer when you ask. AI agents are different. They can plan and carry out multi-step tasks on their own, like a junior co-worker who can follow instructions and check several places for information.

The Financial Times reports that many leaders are still figuring out how to bring these agents into daily work without upsetting staff. At cyber security company Upstream Security, the CEO says people went quiet for about a day after an agent joined a marketing Slack chat. Then someone finally said hello, and the agent started participating. Upstream later expanded agents to other groups for routine tasks that used to take a long time, especially work that involves pulling data from multiple sources.

Research from KPMG suggests adoption is still limited. Only 9 percent of businesses say they have active agents across different parts of their organisation. Around a fifth are exploring the idea, and 17 percent are running pilots.

Some large employers are pushing ahead. Adecco, a global recruitment company, says it is rolling out agents as part of a plan to have 50 percent of revenue “powered by agentic AI” by the end of this year. In the UK, Adecco used an agent for early conversations with job candidates. It found about half of those chats happened outside normal working hours, mostly between 11pm and 3am, and it says the process has saved 20 percent of time so far.

What to watch

Cost is a major issue. Many AI systems charge per “token” (a small unit of text processing, like paying for minutes on a phone plan). Because agents can keep working in the background, companies may need stricter rules, or fixed-price plans, to avoid surprise bills. Managers will also need to plan for more checking and review, since faster work can create more follow-up work for humans.

Source: Financial Times