Deloitte published survey results on August 12 examining how far enterprises have actually gotten with agentic AI, and the gap between testing and true scale is the headline finding. Of 501 senior manager-to-C-suite respondents across five industries in the United States, all from organizations already piloting agentic AI, 42% say their organization has tested or deployed AI agents, and 43% have deployed them across more than one function. But only 15% report having scaled, orchestrated, multi-agent adoption actually in place, and Deloitte notes that many of those deployments remain concentrated in low-risk, low-return applications rather than the higher-stakes workflows where agentic AI’s business case is strongest.
How the survey was built
The survey was fielded between April and June 2026 across five industries: Consumer; Energy, Resources and Industrials; Financial Services; Life Sciences and Health Care; and Technology, Media and Telecom. Respondents split roughly 55% technology functional leaders and 45% business or other functional leaders, and Deloitte augmented the quantitative results with 20 executive interviews. China Widener, a Deloitte vice chair, summarized the core finding directly: “The value in agentic AI depends on more than the agents alone.”
Deloitte deliberately restricted the sample to organizations already piloting agentic AI, which matters for how the numbers should be read. This is not a survey measuring whether enterprises are interested in AI agents in the abstract; every respondent’s employer had already cleared the first hurdle of getting a pilot funded and running. That framing makes the 42%-to-15% drop-off a stronger signal than it would be in a general population survey, since it describes what happens after an organization has already committed budget and executive attention to agentic AI, not what happens before.
Why the gap exists
The pattern Deloitte describes, broad testing but narrow scaling, tracks with what enterprise AI buyers have been reporting anecdotally through 2026: agents are relatively easy to pilot inside a single function or workflow, and considerably harder to orchestrate across functions with the governance, monitoring, and handoff logic that production-grade, multi-agent systems require. The 42%-to-15% drop-off is a specific, sourced number for something most enterprise AI leaders already suspected: pilot success does not predict production success, and the two numbers should be tracked separately by any organization reporting its own agentic AI progress internally.
What this means for the enterprise AI buying committee
Three practical takeaways follow directly from Deloitte’s data:
- Treat “42% have tested agents” as a low bar, not a benchmark to beat. With over 4 in 10 organizations already at that stage, testing alone is no longer a differentiator; the real competitive question is whether an organization is on track to join the 15% that have scaled.
- Push pilots toward higher-risk, higher-return use cases deliberately. Deloitte’s finding that many scaled deployments remain in low-risk applications suggests organizations are optimizing for easy wins rather than the use cases where agentic AI actually changes unit economics. A buying committee should ask its own AI leadership which category its current pilots fall into.
- Budget for orchestration, not just agents. The gap between deployment and scale is explicitly an orchestration and governance problem in Deloitte’s framing, not a model-capability problem. Vendor evaluations and internal build decisions should weight orchestration tooling accordingly rather than assuming agent quality alone determines outcomes.
The read this survey doesn’t spell out
A 27-point gap between organizations that have tested agents and organizations that have scaled them, drawn from a sample where every respondent’s organization was already piloting agentic AI, suggests the industry conversation has moved past “will enterprises adopt agents” and into a much narrower, harder question: which specific organizational capabilities separate the 15% that scale from the 27% that stall after testing. Deloitte’s own emphasis on value depending on “more than the agents alone” points toward governance and orchestration as that differentiator, a claim enterprise buyers should test directly against their own AI vendors’ orchestration capabilities rather than accepting agent-quality marketing alone.
Source: Deloitte Press Room
