OutSystems warns of AI agent governance readiness gap

OutSystems has published research highlighting a gap between executive confidence in governing AI agents and operational readiness to manage them. The study found that only 26% of enterprises say they are well prepared to resolve conflicts across agents, systems and automated workflows.

The findings are based on a survey of 858 executives conducted with CIO Dive and KPMG. They point to a divide between broad confidence in oversight and the practical systems needed as AI agents move into production.

According to the research, 82% of executives believe their organisations can keep pace with governing, integrating and monitoring AI agents. But that figure falls sharply when respondents are asked about resolving conflicts between agents, applications and workflows.

The data suggests many companies are relying on existing governance structures rather than dedicated controls for agentic AI. It found that 83% use current IT, security or compliance processes, specialised reviews, or centralised policies and monitoring to oversee agents that can take autonomous action across applications or workflows.

Only 5% said they have automated controls, audit trails and formal human review points in place, leaving most oversight dependent on manual checks and procedural controls.

Operational gap

The report argues that the challenge for companies is shifting from building individual AI agents to managing broader systems that include models, applications, data sources and third-party tools. It also suggests buyers are trying to avoid heavy dependence on a single supplier.

Among those surveyed, 74% said integration with existing applications and systems was a priority. A further 63% pointed to agent orchestration across systems, while 62% said it was important to retain the option to switch models, vendors and platforms over time.

Testing across full workflows also emerged as a central concern. The research found that 81% of executives consider testing, evaluating and validating agent behaviour very or extremely important when selecting AI platforms or tools.

That focus extends beyond the behaviour of a single agent. Companies are paying attention to agent-to-agent interactions, exception handling, conflicting actions and rollback mechanisms, reflecting concern about how these systems behave in more complex operating environments.

Data consistency is another issue. More than one-third of respondents, or 35%, identified data consistency and context sharing as a major orchestration hurdle, underlining the difficulty of ensuring agents act on governed business data and application states rather than generic model outputs.

"82% of executives say they can keep pace with governing AI agents, yet only 26% say they can resolve conflicts between them," said Tiago Azevedo, Chief Information Officer at OutSystems.

"That gap is where the real work happens. Having put agents to work across the business, I've learned that safety comes from building control into the platform, not reviewing it afterward. Enterprises that enforce policy while agents run will move faster, because they won't have to stop and ask if they can trust what they built," Azevedo said.

Regional picture

The survey included 209 respondents from Australia, India and Japan, offering a view of Asia-Pacific attitudes alongside Europe and North America. On several measures, Asia-Pacific reported stronger preparedness.

In the region, 39% said they were very well prepared to resolve AI agent conflicts, compared with 29% in Europe and 18% in North America. Asia-Pacific also led on automated governance controls, with 14% reporting automated controls, audit trails and human review points, versus 3% in Europe and 2% in North America.

The report also found differences in how organisations assess the value of agentic AI. Globally, 64% cited error reduction as the leading impact metric, followed by compliance and risk reduction, employee productivity and cost reduction.

Employee productivity carried more weight in Asia-Pacific than elsewhere. Some 73% of respondents in Australia, India and Japan used productivity as a measure of impact, compared with 52% in Europe and 45% in North America.

At the same time, the study suggests broader commercial transformation remains limited. Just 12% of respondents globally said agentic AI had accelerated the launch of new products, services or customer experiences.

KPMG said the findings show existing software governance models are not enough for dynamic AI systems. The issue is less whether companies have policies than whether those policies can be applied consistently while agents are operating.

"Assurance for agentic systems must be engineered from day one-you cannot simply inspect it after the fact," said Michael Harper, Managing Director at KPMG.

"Traditional logging and procedural governance were built for static software, not dynamic AI. To bridge the gap between high confidence and true operational readiness, enterprises need the ability to execute policy-as-code and runtime guardrails. That gives businesses the autonomy to innovate while keeping end-to-end workflows secure, reliable and focused on business outcomes," Harper said.