Agentic artificial intelligence (AI) is beginning to move into federal operations, but new MeriTalk research suggests the harder challenge for agencies will not be deploying the technology, it will be redesigning the work around it.
A recent report from MeriTalk and ICF, The Agentic Future: How Federal Agencies Are Redesigning Workflows for the Next Era of AI, finds that 45% of federal IT leaders say their organizations are already using agentic AI in at least one production or operational workflow, while another 34% are exploring or piloting the technology.
Yet only 10% consider agentic AI operationally relevant to their organizations today. Looking ahead, that picture changes dramatically: 85% expect agentic AI to become critical or very important to their missions over the next three years.
The report is based on a survey of 100 federal IT decision-makers from civilian, defense, and national security agencies.
The findings point to an important shift in the federal AI conversation. The challenge is increasingly moving beyond whether agencies can deploy AI to whether their workflows, governance structures, data environments, and employees are ready for systems capable of taking on more complex tasks.
Federal IT leaders recognize that challenge: 91% say redesigning workflows will be a bigger challenge than deploying agentic AI technology.
The research also uncovers a significant gap between perceived readiness and action. While 94% say their organizations are prepared from a security and oversight perspective, only 35% have defined human review or escalation rules. Similarly, 92% report being prepared on governance and policy, but just 34% have clarified decision authority and accountability.
The pattern extends elsewhere. Ninety-two percent report workforce readiness, but only 33% have trained staff on agentic AI. And while 87% say their systems and integration are prepared, just 30% have improved integration, APIs, or workflow orchestration.
Those gaps matter because agentic AI asks agencies to decide not only what an AI system can do, but also what it should be allowed to do, which systems and data it can access, when humans must intervene, and who ultimately owns the outcome.
Federal IT leaders are already identifying the guardrails they will need. Fifty-one percent want identity and access controls specific to AI agents before allowing them to coordinate work across multiple systems or teams. Forty-two percent cite incident response processes for workflow failures, while the same percentage want human approval thresholds based on risk, cost, or public impact.
Agentic AI could also reshape the federal workforce. Nearly all respondents – 98% – expect it to change the role of federal employees. Sixty percent anticipate a greater need for data, AI, or workflow literacy, 59% expect employees to spend more time validating AI-generated outputs, and 56% foresee more time supervising automated workflows.
Early adopters are already reporting measurable benefits. Among respondents actively using, exploring, or piloting agentic AI, 62% cite benefits in software development, testing, or code review; 57% in data validation, reconciliation, or quality review; and 56% each in compliance and audit documentation and cyber threat detection, investigation, or incident response.
But scaling those gains will require organizational change. The main obstacle respondents identify is an inability to align internal teams, cited by 42%. More than half – 52% – say agencies need better coordination across IT, mission, data, cyber, legal, and procurement teams.
The findings suggest that the agentic AI transition will be as much an operating-model challenge as a technology transition. Agencies have begun putting the technology to work, but moving from experimentation to trusted mission operations will depend on modern systems, clear accountability, reliable data, trained employees, and well-defined human oversight.