Stack Overflow has published research showing that workplace use of AI agents has risen sharply over the past year, though most organisations still limit those systems' autonomy.
The findings are based on responses from 1,100 developers and technology professionals across 177 countries. They show growing use of AI agents in day-to-day work, while human oversight remains the norm.
Workplace AI agent use rose to 59% from 31% a year earlier, according to the survey. Daily use increased to 37% from 14%, suggesting the tools are moving beyond occasional experimentation into regular workflows.
Even so, businesses remain reluctant to let agents act independently. Nearly two-thirds of respondents, or 63%, said they rarely or never allow agents to complete tasks without human intervention, while 60% said agents are blocked from making unapproved system changes.
Accuracy concerns
Concerns about reliability were the biggest brake on wider autonomy. More than eight in 10 respondents, or 82%, said they were concerned about the accuracy of information provided by AI agents, while 77% cited security and data privacy.
That caution appears to be shaping how companies deploy the technology. Rather than adopting more complex arrangements, 68% said they preferred predictable single-agent setups over multi-agent systems.
The data also showed a gap between senior leadership and other groups in how often AI agents are used. Senior executives were among the most frequent users, with half reporting daily use, alongside 52% of software architects.
By contrast, enthusiasm at the top has not yet translated into broad confidence in unsupervised deployment. Businesses appear more willing to introduce agentic tools into workflows than to let them act without approval.
Cost shifts
Cost was less of a barrier than a year earlier. Some 38% of respondents said cost was an obstacle to adoption, down from 53% in the previous survey period.
Views on spending differed sharply by group. Senior executives were the least likely to say cost was an issue, at 24%, compared with 60% of students.
The figures suggest budget concerns are easing even as trust and governance questions remain unresolved. For many organisations, the obstacle is no longer whether to spend on AI agents, but how much independence to give them once deployed.
Sector variation
Use of AI agents also varied by industry. Fintech recorded the highest level of daily use at 55%, followed by media and advertising at 50% and software development at 44%.
Those sectors tend to be early adopters of automation and data-heavy workflows, making them a useful indicator of how agent-based tools are being tested in practice. The spread between industries also suggests uptake is not yet uniform across the wider economy.
Overall, the survey paints a picture of a workplace where AI agents are becoming common but remain tightly supervised. Adoption may have nearly doubled in a year, yet the prevailing model is still one of constrained use, close review and limited authority.