Enterprises are wide awake on agentic AI—but need help navigating the chaos

Agentic is charging hard and fast at enterprise leaders and you are worried that support for its responsible adoption is lagging.

The latest HFS OneCouncil session brought enterprise leaders together to unpack their reality of agentic AI to date. The virtual session convened leaders from banking, insurance, healthcare, telecoms, public sector, and investment management to explore their progress with agentic AI. While perspectives varied, the majority in the virtual room represented the banking and financial services (BFS) community, suggesting keen interest as the sector wrestles with agent adoption.

Appetite is firm—but definitions remain confused and confusing

The group broadly agreed: ‘Agentic AI’ isn’t just hype—it represents a significant shift in how enterprises automate and orchestrate work (see Exhibit 1). But, with no consistent definition, many are struggling to make the case internally. As one leader put it; “Even if we define agentic AI today, that definition will probably change in three months.”

Exhibit 1: HFS offers our definition of agentic AI as part of an AI-continuum—a virtual co-worker able to understand goals set by humans—and figuring out how to achieve them

Source: HFS Research, 2025

What’s clear is that agentic AI typically blends LLMs with code to deliver goal-driven, autonomous action. This control over inputs and outputs is attractive to enterprises—but only if it can be governed safely and deployed meaningfully.

Most leaders in our OneCouncil session said they were still early in their journey. A few shared that they were piloting agent use cases, especially in enterprise automation. But others are still trying to align leadership understanding and risk appetite before moving forward. All recognize that the speed of innovation is outpacing the ability to manage it.

The big issues enterprises are calling out: governance, readiness, and value

1. Risk frameworks lag behind ambition

Highly regulated firms remain cautious. Agentic AI’s potential is clear, but without shared definitions, enterprises can’t set meaningful governance standards. Risk teams are understandably wary—especially where agents may interact with customers.

2. Infrastructure is a major blocker

Legacy systems and fragmented data are the top reasons agentic AI isn’t scaling. Connecting LLMs to real enterprise workflows demands integration maturity many firms haven’t yet achieved.

3. Talent strategy and organizational design remain unresolved

Agentic AI is triggering serious questions about performance, accountability, and team structure. Do we reward individuals based on what their agents achieve? What if agents fail? Some leaders highlighted the need to rethink hiring and upskilling in light of synthetic labor.

4. Governance is fragmented and duplicative

Several leaders flagged concerns about duplication of effort within their own organizations—different teams developing agents without visibility or control. There’s growing demand for centralized governance models or transformation offices that can align and rationalize experimentation.

5. Value needs to move beyond headcount

Most early success stories are in back-office automation, boosting speed and customer experience. But efficiency remains the dominant narrative—few have yet shifted the conversation toward growth or innovation. This risks limiting the strategic value of agentic AI.

Leaders want more clarity and need more support

Enterprises are not short on interest—they’re short on clarity and support. OneCouncil members are calling out for:

• A clear, common language to describe agentic AI and its components
• Playbooks to help match the right tools to the right problems
• Governance frameworks and risk models—especially for regulated environments
• Use case libraries with measurable outcomes
• Insights on workforce impact and the emergence of synthetic labor
• Transparency on failures, not just successes—so others can avoid repeating mistakes

Agent connectors gain traction amid rising demand for interoperability

Perhaps because leaders have experienced the consequences of disconnected applications proliferating across their enterprises, there is growing attention on interoperability in the world of agentic AI.

Several leaders referenced Model Context Protocol (MCP – originally developed by Anthropic), Agent-2-Agent (A2A -Google), and the open-agent ecosystem initiatives led by IBM and MIT Media Lab. These efforts are gaining traction, but enterprises need help understanding what they mean in practice.

Now, the market must respond to enterprise concerns

The ecosystem of firms serving enterprise needs for agentic AI must move beyond cheerleading and start solving. Based on our initial read of current enterprise needs, that means an urgent requirement for:

• Standardized terminology and frameworks to help enterprises align stakeholders
• Honesty about failures as lessons learned and shared
• Integration enablement—none of this will work without clean data and working APIs
• Outcomes beyond cost takeout: Measurable CX improvement, speed to market, and operational resilience
• Support for HR and workforce planning with insights into synthetic labor strategy
• Open ecosystems to avoid proprietary lock-ins
• Providers who shift their focus from selling capability to enabling outcomes and who share evidence of real-world deployments, not aspirational roadmaps

The Bottom Line: Enterprise leaders buy the value agentic AI can bring, but they need more help to make it safe and scalable.

Agentic AI is on the move. Enterprise leaders across sectors are engaged and curious but constrained by risk, systems, and uncertainty. They’re not asking whether agentic matters. They’re asking how to do it safely, sensibly, and at scale. Services firms: it’s time to get practical.

• If you’re an enterprise leader seeking honest, peer-led discussion and practical insight into the future of AI and other business challenges, HFS OneCouncil is open to you. Learn more here: https://www.hfsresearch.com/hfs-onecouncil/