A September 2025 survey of 100 US healthcare technology executives gives the clearest picture yet of agentic AI adoption. The numbers are strong, and the sample is narrower than the headlines suggest.
In a Deloitte survey of 100 US healthcare technology executives conducted in September 2025, 61% said they were already building and implementing agentic AI initiatives or had secured budgets, and 85% planned to increase investment over the next two to three years. Expectations on return are high: 98% expect at least 10% cost savings in that timeframe. The sample is 50 health systems and 50 health plans, all with revenues above 500 million US dollars, so the figures describe large, well-resourced US organisations rather than healthcare generally.
Deloitte surveyed 100 US healthcare technology executives in September 2025: chief technology, information, digital, AI and innovation officers, plus vice presidents of technology, digital and strategy. The sample splits evenly, 50 from health systems and 50 from health plans, all at organisations with annual revenues above 500 million US dollars. A further 35 executives and agentic AI leaders took part in focus groups.
The headline figures:
Executives reported that the obstacles which stalled earlier waves of healthcare AI have softened. 40% said technical talent limitations were no longer a major challenge. 38% reported reduced resistance to change. 35% reported improved leadership buy-in. 32% reported reduced concerns about data quality.
Read together, this describes a shift in what is holding deployment back. The constraint is moving away from whether the organisation can build the thing, and towards whether it can decide what is worth building.
The survey splits respondents by posture, and the two groups make structurally different bets. 82% of early adopters prioritise multi-agent solutions. 92% of watchers prefer point solutions. The expectations diverge sharply too: 59% of early adopters expect savings above 20%, against 13% of watchers.
Whether that reflects better information or more optimism is not something the survey can settle.
Three limits worth holding onto. The sample is 100 executives at large US organisations, so it does not describe smaller providers or non-US systems, and nothing in it speaks to Asia Pacific. The respondents are technology leaders, who are not neutral observers of technology budgets. And the return figures are expectations, gathered before most of these deployments matured, not measured outcomes.
We checked several widely circulated figures on agentic AI in healthcare while writing this, and most traced back to vendor marketing pages rather than to a survey with a stated methodology. One commonly quoted claim, that 68% of healthcare organisations are currently using AI agents, we could not tie to any primary source. Deloitte's 61% is a different and more carefully bounded statement: already building, implementing, or budget secured, among 100 large US organisations. If you are putting a number in a board paper, use one you can trace.
On the substance, the pattern we see with clients matches the survey's direction. Agents land first where the task is repetitive, rules-heavy and administratively painful. Prior authorisation, documentation, scheduling, revenue cycle. Not diagnosis. The organisations getting value are not the ones with the most ambitious clinical pilots, they are the ones that picked an unglamorous workflow with a measurable baseline.
The finding we would act on is that 98% expecting at least 10% cost savings. Expectations that uniform are rarely all correct. Some of those programmes will underdeliver, and the ones that do will mostly be the ones that never established what the process cost before the agent arrived. Measure the baseline first. It is the cheapest insurance available.