People are handing work to AI agents even though they are unreliable and lack accountability

UK workers expect to hand over almost a third of their workload to AI agents within the next year, yet most say the technology is still unreliable and lacks clear accountability, according to a new poll from Asana’s Work Innovation Lab. The survey of more than 2,000 workers, including 1,021 in the UK, found that employees anticipate delegating 32 percent of tasks to AI within 12 months and 41 percent within three years. At present, only a quarter feel ready to do so. Despite the rapid spread of adoption, with nearly three-quarters already using AI agents, concerns about reliability and oversight remain high.

Almost two-thirds of workers believe AI agents are unreliable, with more than half saying they ignore feedback or provide incorrect information. Only 18 percent of organisations currently measure errors, despite 63 percent of employees citing accuracy as the most important performance metric. Fewer than one in ten companies have established ethical frameworks or review processes, and just 12 percent have clear guidelines on which tasks should remain the responsibility of humans.

The research highlights a growing accountability gap. When AI agents make mistakes, blame is divided between end users, IT teams and creators, but over a third of employees say they do not know who should take responsibility. This lack of clarity risks leaving errors unchecked and eroding trust.

Training remains another barrier. While 82 percent of workers want guidance on how to use AI agents effectively, only 32 percent say their organisation has provided it. Many employees are also calling for clearer rules, with more than half asking for defined boundaries between human and AI responsibilities.

Despite these challenges, the majority of workers see AI agents as a fundamental shift in how work is organised rather than a passing productivity tool. Employees most often use them for tasks such as organising or finding documents and scheduling meetings. Many even prefer delegating these activities to machines rather than colleagues.

Mark Hoffman, Work Innovation Lead at Asana, said the findings show that organisations need to treat AI agents like teammates rather than tools. He argues this means defining responsibilities, embedding feedback loops, and equipping employees with the training and context to use them effectively. Without such measures, the report warns that companies risk accumulating “AI debt” in the form of unreliable systems and weak oversight.