A supervisor oversees a group of agents while a larger network remains outside oversight and shows alerts.

AI agents: companies are adopting them faster than they can oversee them

AI agents are no longer an experiment: companies are moving them into production. But the ability to oversee them is advancing more slowly than their adoption. That is the conclusion of a study by OutSystems, CIO Dive and KPMG published on October 16, 2026, based on a survey of 858 technology executives (OutSystems, 2026).

The gap between confidence and readiness

  • High confidence: 82% believe they will be able to keep pace in governing their AI agents.
  • Low readiness: only 26% say they are well prepared to resolve conflicts between agents, systems and automated workflows.
  • Minimal controls: 83% oversee their agents with the governance processes they already had, and only 5% have automated controls, audit trails and formal human review checkpoints.

According to the study, 81% consider it very or extremely important to test and validate agent behavior before choosing a platform, and 35% cite the lack of consistent data and shared context as a major obstacle.

Where the value shows (and where it doesn’t yet)

The most frequently cited impact metric is error reduction, mentioned by 64% of respondents. By contrast, only 12% say agents have already accelerated new products, services or customer experiences. The trend repeats across the region: in Peru, an Infobae analysis warns that companies are adding agents faster than they can oversee them and measure their return (Infobae, 2026, in Spanish).

What does it mean for your business?

  • Define what each agent can do: least-privilege permissions and human approval for sensitive actions, such as payments, messages to customers or changes to systems.
  • Log everything: every agent action should leave a trace so it can be reviewed and corrected.
  • Test before you scale: validate the agent with real cases and define when it should hand the case over to a person.
  • Measure from the start: time, errors and cost, before and after.
  • Get your data in order: an agent decides only as well as the information it consults.

At Wasivia, we design AI agents and multi-agent systems with least-privilege permissions, human oversight and a log of every action. Read also Shadow AI and agents: the new AI security risks in the enterprise. Book a 30-minute assessment.

References

Similar Posts