A survey of 300 technology leaders, developers and artificial intelligence (AI) practitioners published today finds 72% of respondents reporting their organization is actively using agentic AI systems, with another 21% of respondents planning to implement agentic AI systems within the next 24 months.
Conducted by Gravitee, the provider of a platform for managing application programming interfaces (APIs), the survey says nearly half of the respondents (49%) reported that their initiatives are backed by a net new budget specifically allocated for agentic AI, while 36% are reallocating from existing budgets, without cutting into other IT initiatives. Well over a third (38%) work for organizations that have established a dedicated agentic AI team to achieve that goal.
The largest group of respondents (49%) reported annual spends between $50,000 and $249,999 on large language models (LLMs), and full 89% expect to increase their spending on LLMs, the survey finds.
The primary justification for making those agentic AI investments is for achieving operational efficiency (74%), followed by improving customer experience (46%) and reducing cost (38%). A total of 43% identified automating some type of task or process as their top use case, with chatbots (20%) a close second.
The degree to which organizations will achieve those goals will depend heavily on how well AI agents are integrated, but it’s apparent that organizations are trying to seize the opportunity, says Gravitee CEO Rory Blundell. “This is happening at a very large scale,” he adds.
The challenge is not just building AI agents but also making them aware of each to the point where a task spanning, for example, sales, marketing and manufacturing, can be automated, Blundell says. Other major challenges include controlling costs (45%) and flooding large language models with too many requests (23%). Three quarters of respondents (75%) also identified governance as a high priority.
Overall, OpenAI is the most widely used AI platform (49%), followed by Google Vertex AI (10%), Microsoft Azure (9%), and IBM (9%) also showing strong traction. OpenAI’s ChatGPT is the most widely used entry point for organizations adopting large language models (LLMs), with 87% of respondents indicating past usage.
However, LLMs from Google (58%) and Deepseek (27%) are also gaining traction, which suggests organizations are now using multiple LLMs. Nearly two thirds of the respondents expect to use multiple LLMs, with 45% planning to use three or more.
However, 97% plan to standardize on a single provider that provides access to multiple LLMs, the survey finds.
At this juncture, it’s not so much a question of whether AI agents will be used but rather to what extent. Generative AI is probabilistic in the sense that it surfaces the best guess about what should come next in a sequence. That works well when crafting a marketing document or writing code, but applying that to tasks requires more advanced reasoning capabilities. Additionally, generative AI platforms employed today rarely perform the same task the same way every time, so if a process is deterministic, their applicability might be limited.
On the plus side, however, as models trained using narrow sets of data that relate to one specific function become more available, the overall accuracy of an AI agent should improve.
Ultimately, AI agents should enhance productivity by eliminating much of the toil that is required today to complete a task. It’s not likely, however, they will be able to replace humans any time soon, but the nature of jobs that humans perform will be permanently altered as they find themselves increasingly supervising agents performing a task that previously, if truth be told, they didn’t really enjoy doing themselves in the first place.
