U.S. federal agencies are adopting agentic artificial intelligence (AI), but many still lack the operating and AI governance foundations needed to scale it, according to a new survey released October 19 by ICF, a global solutions and technology provider, and MeriTalk, a government IT news and research platform. 

The research, The Agentic Future: How Federal Agencies Are Redesigning Workflows for the Next Era of AI, surveyed 100 federal IT decision-makers across civilian, defense and national security agencies.

While 45% of respondents’ agencies already use agentic AI in at least one production or operational workflow, and another 34% are exploring or piloting it, only 10% said the technology is operationally relevant across their organizations. 

The report identifies workflow redesign as the biggest obstacle to scaling agentic AI, cited by 91% of respondents, and points to gaps in human oversight, decision authority and employee training.

The findings highlight a growing challenge for chief data officers (CDOs): ensuring AI agents can access information and carry out tasks within clearly defined operational and governance boundaries. 

Oversight and Decision Rights Remain Unclear

The survey found that only 35% of agencies have defined human-in-the-loop review processes for agentic AI. Just 34% have clarified decision authority, while 33% have trained employees to work effectively with AI agents.

These gaps matter as agencies move beyond isolated AI applications toward agents that can coordinate work across multiple systems.

Such deployments require organizations to establish which actions an agent can take independently, when human approval is necessary and who is accountable when an automated decision causes a problem.

Federal IT leaders also identified safeguards needed before agents coordinate work across systems. Agent-specific identity and access controls were cited by 51% of respondents, while 42% identified incident-response processes and 42% pointed to human approval thresholds based on factors such as risk, cost or public impact.

What the Findings Mean for Data Governance

For CDOs, agentic AI introduces a need to govern who can access information and how automated systems use that access to act on it. Permissions designed for employees or conventional applications may not adequately define the boundaries for agents operating across multiple systems.

A robust AI governance strategy can help establish those boundaries through access policies, approval requirements, accountability and monitoring. 

CDOs also need to work with security, technology and business leaders to ensure data access aligns with the purpose and authority granted to each agent.

The need is likely to grow as agencies expand their use of agentic AI. 

In fact, the survey found that 85% of respondents expect agentic AI to become critical to their missions within three years. For CDOs, preparing for that shift means making data access and accountability part of the operating model before agents assume broader responsibilities.