CrewAI Surveyed 500 Enterprise Leaders on AI Agents. Security Ranked First in Platform Selection. ROI Ranked Last.

October 26, 2026 · Josh Agusti, Founder, Aiven Consulting Group

Ask a room of executives whether they plan to keep investing in AI agents this year, and in CrewAI's new 2026 State of Agentic AI survey, every single one of them said yes. That is the easy part. The harder question is what they are actually optimizing for when they pick a system to run those agents, and the answer there is less reassuring than the headline number suggests.

CrewAI surveyed 500 C-level executives and senior leaders at companies with at least $100 million in annual revenue and 5,000 or more employees, spanning seven global regions. The sample is large enough, and senior enough, to treat the result as a real signal about where enterprise priorities actually sit, not just where they claim to sit.

Adoption Is No Longer the Open Question

Sixty-five percent of respondents said their organizations already use AI agents today, and 81 percent describe that use as scaled or expanding rather than stuck in pilot. On average, companies report having automated about 31 percent of their workflows with agents, and they expect that figure to climb by another 33 percent over the course of 2026. The benefits they cite are concrete rather than aspirational: 75 percent point to time savings, 69 percent to lower operational costs, 62 percent to new revenue, and 59 percent to reduced labor costs. Whatever skepticism still existed about agentic AI in the abstract has mostly evaporated inside the companies large enough to have run real pilots.

What They Actually Screen For When Choosing a Platform

The more revealing data sits in how these leaders say they choose a platform once they have decided to move. Security and governance ranked first among selection criteria, cited by 34 percent of respondents, ahead of ease of integration at 30 percent and reliability or performance at 24 percent. Proving time-to-value or return on investment ranked last, at just 2 percent. In other words, the people writing the checks are far more worried about what an agent might do wrong than about whether it will pay for itself. Fifty-seven percent also said they would rather build on tools they already have than adopt something new from scratch, which points toward caution and incrementalism even inside organizations that describe themselves as scaling fast.

The Trust Signal Doesn't Replace the Business Case

Ranking security first is not a bad instinct. It is arguably the correct one, and organizations that have mapped acceptable-use boundaries and escalation paths through something like a formal governance framework are better positioned to evaluate a vendor's claims than ones improvising the standard in the room. But security and governance answer a different question than ROI does. A platform can pass every security review an enterprise throws at it and still fail to move a single meaningful metric a year later, because nobody built the measurement discipline to track whether it did.

Liferay's research on enterprise AI agents found a related pattern: more than half of organizations with agents already live in production couldn't measure what those agents were actually doing. That gap in measurement discipline is explored further in Liferay's measurement gap findings, and it mirrors exactly what CrewAI's ranking implies: governance gets attention, outcomes measurement does not. Security review and ROI measurement are not substitutes for each other. Treating the first as sufficient is how a program ends up fully governed, widely deployed, and still unable to answer the one question that eventually determines its budget.

Build the Business Case Before the Rollout, Not After

The fix is not complicated, it is just rarely sequenced correctly. Before expanding agent use further, it is worth running the numbers on what a given deployment is supposed to return, using something as basic as an ROI calculator to force the question into the open before the rollout rather than a year into it. CrewAI's 100 percent expansion number is a sign of real momentum. Whether that momentum compounds into value or just into more deployed infrastructure depends on whether the measurement discipline catches up to the governance discipline, and right now, for most of the leaders in this survey, it has not.

Key Takeaways

CrewAI's 2026 State of Agentic AI survey of 500 enterprise leaders found 100% plan to expand AI agent use in 2026, with 65% already using agents in production and 81% describing that use as scaled or expanding.
When choosing an AI agent platform, security and governance ranked as the top selection criterion at 34%, while proving return on investment or time-to-value ranked last at just 2%.
The mismatch is a warning sign: organizations are getting serious about governing agents but not about measuring whether they deliver value, which leaves budgets exposed once leadership asks for proof of return.