60% of AI agents are embedded in IT departments – here's what they're doing

Research suggests almost everyone wants an AI agent — they’re the coolest thing since sliced data. But what exactly are these agents doing within enterprises? In many cases, their job may be to help build even more agents. In most instances, agents help IT departments manage system performance, presumably including the technical underpinnings of AI agents. However, use cases vary by industry. 

A mind-blowing 96% of organizations plan to expand their use of AI during the next 12 months, according to a recent survey of 1,484 IT leaders from technology specialist Cloudera. That’s a huge percentage for any survey topic — a minimum of 10% of respondents are usually outliers. A majority, 57%, said they’ve already implemented AI agents in the past two years. At the same time, fears around data privacy, integration, and data quality may potentially spoil the party, the survey suggests.

Also: The 4 types of people interested in AI agents – and what businesses can learn from them

Most (61%) AI agents in production are embedded within IT operations. The leading applications being assumed by agents include performance optimization bots (66%), security monitoring agents (63%), and development assistants (62%).

So, where are these agents coming from? Two-thirds of respondents (66%) build agents on enterprise AI infrastructure platforms, while 60% leverage agentic capabilities embedded in core applications. “This hybrid approach reflects a clear preference for scalable, secure, and close-to-data deployments,” the survey’s authors said.

Outside of IT optimization, early deployments of AI agents tend to focus on customer-facing operations. AI agents are most used for customer support (78%), process automation (71%), and predictive analytics (57%). 

Also: Why neglecting AI ethics is such risky business – and how to do AI right

When asked which technologies they currently use or plan to use to build agentic AI, respondents identified enterprise AI infrastructure platforms (66%), agent capabilities embedded within applications (60%), and dedicated enterprise AI agent platforms and frameworks (60%). 

AI agents aren’t perfect, of course, and deployers encounter many of the same issues as previous generations of technology. Top concerns with AI agent deployments include data privacy concerns (53%), integration with existing systems (40%), and high implementation costs (39%). 

More than a third (37%) of respondents report that integrating AI agents into current systems and workflows has been “very” or “extremely” challenging. “In other words, deploying AI agents is not a plug-and-play endeavor,” the authors said. Again, the more things change with technology, the more the challenges remain the same.  

Agentic AI vendors and proponents looking to push further change have their work cut out. Technology leaders would like to see more features in the AI agents they deploy, including stronger data privacy and security features (65%), faster training and customization (54%), enhanced natural language processing (51%), and better contextual understanding (50%).

Also: How an ‘internet of agents’ could help AIs connect and work together

You can add to the list of challenges a wide variety of use cases by industry, including:

• In finance and insurance, fraud detection (56%), risk assessment (44%), and investment advisory (38%) are the leading use cases. 
• In manufacturing, top applications include process automation (49%), supply chain optimization (48%), and quality control (47%). 
• In healthcare, leading use cases include appointment scheduling (51%), diagnostic assistance (50%), and medical records processing (47%).
• In telecommunications, top applications are customer support bots (49%), customer experience agents (44%), and security monitoring agents (49%).

The Cloudera authors made recommendations for implementing AI agents. 

Also: How AI will transform cybersecurity in 2025 – and supercharge cybercrime

Once again, for the most part, these approaches are not novel or new to those who have implemented previous generations of technology:

• Strengthen data foundation and integration capabilities: “Enterprises should ensure they have modern data architecture and unified platforms that can securely handle the volume and variety of data that AI agents require.”
• Start with high-impact projects to deliver immediate ROI and grow from there: “Survey respondents focused on customer support and process automation as initial use cases, suggesting these areas are good launch pads because they address real pain points and have measurable outcomes.”
• Establish accountability: “Enterprises must clarify: who is responsible for an agent’s performance? Is it the developer who built it, the business owner who uses it, or the operations team that oversees it?”
• Build governance and ethics frameworks: “Include mechanisms to audit bias, ensure transparency in agent decision-making, and regularly review agent behavior against enterprise policies and user expectations.”
• Upskill teams and foster a culture of human-AI collaboration: Go “beyond basic training to cultivate hybrid skill sets — people who can not only build and integrate AI agents, but also understand their reasoning, limitations, and evolving capabilities. Prioritize hands-on, continuous learning, encouraging experimentation and knowledge-sharing across roles.”

The strength of feeling in the survey responses suggests that AI agents are the next wave of AI, providing focused initiatives for specific functions versus the huge, complicated AI systems that many business leaders were dreading. It will be interesting to see if that 96% planned adoption rate holds.