Most Enterprises Have Deployed AI Agents But Few Trust Them, Forrester Research Finds

The vast majority of enterprises have moved AI agents into production, but confidence in what those agents are actually doing remains low, according to new research commissioned by data platform vendor Boomi and conducted by Forrester Consulting.

The study surveyed 409 IT and technology decision-makers at director level and above across North America, Europe and APAC. It found that while 86 per cent of organisations have progressed beyond the AI agent pilot stage, only 34 per cent trust the actions their AI agents are taking.

The gap between deployment velocity and operational readiness is creating what the research describes as “agentic chaos” – a state in which organisations push AI agents into production without the governance, integration or infrastructure to support them.

Integration, Not Model Quality, Is The Dividing Line

The Forrester research identifies integration as the primary factor separating enterprises that trust their AI agents from those that do not.

Organisations classified as having “agentic control” – the top quartile for operational readiness across governance, integration, API and MCP management and other categories – were three times as likely as those in agentic chaos to say that reliable, well-managed APIs determine whether they pilot a use case at all.

The adoption of integration platform as a service (iPaaS) showed the widest gap of any method surveyed. Among organisations with agentic control, 46 per cent use iPaaS to support agentic workflows. In the bottom quartile, that figure drops to 25 per cent.

The difference is even more pronounced when it comes to building agents. Eighty-six per cent of organisations with agentic control consider iPaaS and API management capabilities important to their readiness, compared with 58 per cent of those in agentic chaos. Forrester identified this as the single widest gap measured in the entire study.

Steve Lucas, Chairman and CEO at Boomi, framed the findings as a data infrastructure problem rather than a model capability problem.

“The trust problem with agentic AI is really a data problem,” Lucas noted. “Agents can only be trusted to act on data that’s been properly activated, connected and governed, and most companies deployed agents before they did that work. The ones who did it first are the ones getting real value now.”

Deploying Without Readiness Carries A Cost

Among organisations in the bottom quartile for readiness, 77 per cent are moving into production regardless, exposing themselves to what the research estimates is an average of $2.1 million in added costs. Those costs stem from compliance fines, lost customers, operational downtime and rework.

By contrast, organisations with agentic control reported far higher confidence in their deployments. Fifty-five per cent of those in the top quartile indicated high confidence in their agents’ actions and decisions, compared with just 22 per cent of those in agentic chaos.

The research also found that organisations with agentic control are more likely to prioritise the operational work that sits behind the AI layer. Forty-seven per cent cited improving integration with tools, APIs and apps as a focus area, versus 31 per cent of those in chaos, who remain focused on improving the AI model itself without building the connections needed to act on its outputs.

Agent Sprawl Is Becoming A Problem

Some organisations are now running as many as 200 agents, a symptom of what Forrester describes as “POC/pilot purgatory” – a pattern where enthusiasm for AI agents stalls because those agents were never connected to the enterprise systems they need to act on their decisions.

Organisations with agentic control are more likely to have taken steps to manage this sprawl. Forty-six per cent have established centralised governance of MCP (Model Context Protocol), the standard that allows agents to connect to enterprise systems, compared with 32 per cent of those in chaos.

Forty-eight per cent of those with agentic control have aligned their AI and integration teams under a single operating model, compared with 37 per cent of those still in agentic chaos.

Forrester’s characterisation of iPaaS in the study drew a direct analogy to the role integration plays in making AI useful: “If LLMs are like brains, iPaaS products are a limb that empowers that brain to act upon the physical world.”

Where The Returns Are Showing Up

The organisations that have invested in integration infrastructure are reporting measurable returns from their agentic AI deployments. Among those with agentic control, 59 per cent reported productivity gains, 51 per cent reported increased innovation, 46 per cent gained reusable capabilities they could apply elsewhere in the business and 45 per cent automated repetitive tasks.

Forrester’s recommendations for organisations looking to close the trust gap centre on three actions: aligning AI and integration teams under one operating model, implementing a control plane to govern AI agents and introducing an orchestration layer that gives agents the ability to connect to data, applications and each other.

The Broader Implication

The research suggests that the current wave of enterprise AI agent adoption is running ahead of the infrastructure required to make those agents reliable. While the AI models powering agents continue to advance, the study positions integration, governance and API management as the less visible but more consequential factors in determining whether agentic AI delivers on its promise or creates operational risk.

For enterprises already deep into AI agent deployment, the findings point to a practical gap: the organisations seeing returns are not necessarily the ones with the most advanced models, but the ones that built the plumbing first.

The full study, “The Agentic AI Readiness Gap: Insights From Leaders At Organizations Scaling Agentic Pilots To Production,” is available for download from Boomi.

Last Updated on July 21, 2026 by Nick Ross