The AI market poured $581.69 billion into the sector in 2025, a 130 % jump from the previous year and 40 times the 2013 level. Yet scaled use of AI agents remains in the single-digit percent range for almost every business function, according to the Stanford AI Index. The mismatch between capital and rollout is reshaping both vendor strategies and buyer expectations.
Where agents actually run
The Stanford AI Index paints a nuanced picture of adoption:
- 88 % of companies report using AI in at least one area of their operations.
- 70 % say they employ generative AI for at least one function.
- Scaled deployment of AI agents—software that can act autonomously on behalf of users—remains in the single-digit percent range for almost every function.
Only three domains break the single-digit barrier:
| Function | Percentage of firms with scaled agents |
|---|---|
| Software engineering | 24 % |
| IT operations | 22 % |
| Service operations | 21 % |
In finance, marketing, human resources and supply-chain management, agents are still rare, typically confined to experimental pilots.
The survey paradox
Two widely cited surveys appear to contradict each other. One, targeting large enterprises, finds most are still in the pilot phase, testing AI agents without committing to production. The other, aimed at agent engineers, reports that 57 % have agents running in live environments.
Both numbers are accurate; they simply measure different slices of the market. The enterprise survey captures the mass market, where budget approvals, governance concerns and legacy system integration slow rollout. The engineer-focused survey reflects the frontier, where early adopters and specialist teams push agents into production as part of niche projects.
What the gap means for vendors
For companies selling AI agents, the disparity is a clear opening. Buyers no longer want more demos; they want trustworthy, observable, and governable agents that can be handed over to business units as infrastructure rather than proof-of-concepts. Features that manage permissions, audit trails and smooth hand-off between IT and line-of-business teams are becoming decisive.
What buyers should keep in mind
Enterprises should resist the pressure to rush into production simply because the money is flowing. The data shows that most companies have not yet figured out how to move agents beyond pilots. Organizations that take a measured approach—starting with the three functions where adoption is 21-24 % and building strong governance frameworks—will likely reap the greatest benefits.
Key actions for buyers:
- Map pilot projects to a broader rollout plan before allocating additional budget.
- Demand observability tools that surface agent decisions, data usage and error rates.
- Secure clear ownership of agents, separating development, operations and compliance responsibilities.
- Invest in upskilling both IT staff and business users to reduce reliance on external consultants.