.jpg)
How public sector leaders are turning standard operating procedures into scalable agentic workflows
This article is adapted from the first chapter of Distributed AI for the Mission, a recent editorial report published by Corinium and Sun Radia. You can read the full report here.
___
For public sector leaders, AI rarely arrives in a pure environment.
Often, organizations’ data infrastructure is shaped by legacy systems, tight budgets, and strict approval chains, and every modernization decision must be made with public trust in mind.
None of this can be paused while organizations experiment. Leaders must innovate, but they must also protect trust. They must move faster, but they must preserve accountability. They must encourage experimentation, but guard public services.
That makes the challenge of operationalizing AI fundamentally different from proving that AI can work in a controlled setting. A pilot can show that a model can perform discrete tasks or provide information. Production AI must do far more. It must integrate with existing systems, operate on messy data, fit within governance frameworks, support auditability, and earn the confidence of the people expected to use it.
This is why many public sector AI initiatives stall between proof of concept and operational value. The challenge, then, is not whether we can make technology work, it is how to best prepare our organizations to benefit from it.
For Gabe Arrington, Chief, International Affairs for the United States Air Force, the distinction between an interesting AI experiment and an operationally useful capability boils down to three fundamentals: integration, sustainment, and mission alignment.
“The barrier most frequently underestimated is the sustainment tail, specifically machine learning operations and continuous monitoring. Leaders often treat AI like traditional commercial-off-the-shelf technology or software. You buy it, you install it, you move on. In reality, AI models degrade over time as real-world data shifts from training data.”
Gabe Arrington
Chief, International Affairs, United States Air Force
That point is especially important in public sector environments, where AI systems must operate within established controls rather than around them. A technically successful prototype can still fail to create value at scale if it cannot be appropriately monitored, governed, secured, and sustained as conditions change.
Richard Patterson, Chief Data and AI Officer, Bureau of the Comptroller and Global Financial Services for the U.S. Department of State, argues that many AI initiatives struggle because leaders underestimate the full spectrum of support that takes AI from experiment to production scale.
“Someone has a great idea and someone who has some authority likes that idea and they say, let’s go do it. Something gets built and it may not be exactly what the end user needed. It may not fit into the enterprise architecture. It may not have security and governance around the data that needs to be there.”
Richard Patterson
Chief Data and AI Officer, Bureau of the Comptroller and Global Financial Services, U.S. Department of State
The result is a familiar pattern. A team builds something that demonstrates technical promise, but the tool does not align with how work actually gets done. It may depend on data that is inconsistent, poorly governed, or difficult to access.
It may require users to change long-established workflows without giving them enough transparency or incentive to trust the output. It may solve one step in a process while creating friction elsewhere.
In the public sector, these issues are made more pronounced by organizational complexity. Processes often span multiple teams, systems, and approval points. Responsibilities may be distributed across departments.
This means public sector organizations cannot treat AI as a shortcut around foundational work. Scaling AI requires a clear understanding of the mission outcome, the data environment, the enterprise architecture, the decision rights, the risk profile, and the human oversight model.
It also requires a budget and talent strategy for sustainment, because AI systems are not static assets that can be installed and left alone. AI should not be inserted into isolated process steps without considering the wider workflow.
The organizations that move beyond AI pilots reliably will be the ones that connect experimentation to mission outcomes, architecture, governance, workflow design, funding, and change management from the start.
This is the foundation for operational AI. Before agencies can turn standard operating procedures into scalable agentic workflows, they must first close the gap between what AI can do in theory and what public sector organizations need it to do in practice.
Are Your Standard Operating Procedures Ready for Agentic Workflows?
Beyond scaling promising AI use cases, public sector data leaders need to identify workflows that can turn repeatable work into governed workflows that can operate across systems, teams, and approval chains without losing accountability.
That is where standard operating procedures, or SOPs, become critical. In theory, SOPs show leaders where AI can assist, where people must stay in control, and how decisions should move through an organization. In practice, many public sector organizations first have to confront an uncomfortable reality: not every process is as standardized as leaders assume.
According to Richard Patterson, many teams discover this only when preparing a process for automation.
“A lot of the business types of activities, surprisingly, are done through word-of-mouth processes. There might be some training materials, there might be an SOP, it might be 20 years old.”
Richard Patterson
Chief Data and AI Officer, Bureau of the Comptroller and Global Financial Services, U.S. Department of State
This matters because agentic workflows depend on clarity. If a process varies across teams, or exists mainly as institutional knowledge, it becomes difficult to automate responsibly.
Before leaders can decide whether AI should summarize a document, populate a form, trigger an authorization, or escalate a decision, they must understand the real workflow, not just the assumed one.
For Patterson, having a documented SOP is a starting point. From there, leaders can evaluate risk, reward, governance needs, and guardrails. A process with high volume, repeatable steps, and clear escalation points may be a good candidate for AI-enabled orchestration. A process with unclear ownership, inconsistent execution, or high-impact discretionary decisions may need redesign before automation.
Steven Walters, CEO of Property Management Resources LLC, adds that readiness also depends on how an organization defines success up front.
“A workflow is ready for agentic AI when its decision logic is documented, its inputs are clean and governed, and the work has a clear definition of done, an auditable success criterion and a defined exception path. The strongest warning sign is when a process is tribal, executed a little differently by every person. Automating that only scales the chaos. The deepest red flag is intent: using automation as a band-aid to avoid fixing a broken process.”
Steven Walters
CEO, Property Management Resources LLC
Gabe Arrington says leaders should think of AI as a way to orchestrate established procedures, not bypass them. That distinction is especially important in government environments, where approvals, documentation, handoffs, and records often exist for good reasons.
“The goal is to use AI to orchestrate the standard operating procedure, not to bypass it. Accountability is maintained by shifting the human from doer to reviewer. You establish control through robust auditability, every AI action, data retrieval, suggested decision must be logged.”
Gabe Arrington
Chief, International Affairs, United States Air Force
This is the practical promise of agentic AI in public sector environments. Rather than replacing people, agentic systems can coordinate work across fragmented systems and teams.
In doing so, they can reduce the manual “swivel chair” work that slows many government processes, while keeping sensitive decisions under human authority. Arrington describes agentic AI as connective tissue between legacy systems that were never designed to communicate.
This makes workflow orchestration especially relevant for compliance, authorization, data cataloging, and pipeline management. These are areas where work is often structured, documentation-heavy, and dependent on multiple systems.
But orchestration is not just a technical layer. It is also a control mechanism. By designing workflows around audit trails, permission checks, escalation points, and human review, agencies can move faster without weakening governance. This is essential in public sector environments, where speed alone is not enough. AI-enabled workflows must also be explainable, traceable, and legally defensible.
The human role also changes. In an orchestrated workflow, AI can absorb more of the administrative burden, while people focus on higher value tasks.
For Steven Walters, CEO of Property Management Resources LLC, the way to keep human authority meaningful without slowing the process down is to make oversight proportional to risk.
“The way to keep authority meaningful without creating bottlenecks is to make oversight risk-based rather than uniform. Low-risk, repeatable steps run autonomously within clear guardrails, while human review is reserved for consequential judgments and anything that crosses a defined threshold. That’s what turns oversight into a control mechanism embedded in the workflow, not a manual gate bolted onto it.”
Steven Walters
CEO, Property Management Resources LLC
However, the shift only works when users trust the system. Patterson warns that if people do not understand where AI-generated information comes from or how AI-influenced decisions are made, it may damage that trust.
This is why successful AI orchestration must involve end users early. Their job descriptions and performance metrics may also need to evolve so they are not effectively competing with AI for the same task.
Steven Walters, CEO of Property Management Resources LLC, argues that this involvement has to start earlier than most organizations plan for.
“Leaders should involve frontline employees early, during process discovery, not after the workflow is designed. The people doing the work know how a process actually runs versus how it’s assumed to run. That’s also how you address the fear directly: be explicit that the goal is to shift people from doers to reviewers, not to compete with them for their own tasks.”
Steven Walters
CEO, Property Management Resources LLC
Turning SOPs into agentic workflows is therefore not a simple automation exercise. It is a modernization discipline that combines process discovery, governance, enterprise architecture, data management, and change leadership.
This offers a tantalizing opportunity to increase efficiency in the public sector. But the sequence matters. Leaders must first understand the process, then design the workflow, then apply AI within clear boundaries. Only then can agentic AI help public sector organizations scale operational improvements while preserving public trust.
How to Extend Orchestration Across Modernization Initiatives
Once public sector organizations have the necessary foundations for growth, workflow orchestration can support the scaling of automation across the organization in a consistent way.
Many modernization initiatives are not limited by a lack of AI capability. Rather, it’s the fragmented systems, manual handoffs, inconsistent data flows, and processes that lie underneath that slow them down. In these environments, AI may provide insight, but workflow orchestration turns that insight into structured action.
Arrington argues that workflow orchestration is the layer that makes AI operationally useful. Without it, even powerful algorithms remain trapped in isolated systems.
“Workflow orchestration is really the foundational layer that makes AI usable. Without it, you simply have brilliant algorithms created by the greatest minds in the world trapped in isolation.”
Gabe Arrington
Chief, International Affairs, United States Air Force
Patterson expects this model to become increasingly important as AI products become more domain-specific. In areas such as government finance, logistics, compliance, and back-office operations, he sees opportunities for AI-enabled platforms that incorporate common processes and best practices.
“There’s going to be more domain-based products as well, that understand the best practices and the optimal processes to get the most efficient ways out.”
Richard Patterson
Chief Data and AI Officer, Bureau of the Comptroller and Global Financial Services, U.S. Department of State
This is especially important for processes that depend on traceability and repeatable execution. They also require agencies to gather evidence from multiple systems, apply rules, document decisions, and escalate exceptions to the right people.
But the path forward is not simply to buy or build more AI tools. Public sector leaders need to decide where standardized products are appropriate, where workflows should be adapted, and where mission-critical capabilities require custom development. This is where enterprise architecture becomes central to modernization.
Ad hoc experimentation is not enough. If agencies want AI to scale, they must connect it to their future-state architecture, data strategy, governance model, and investment plan.
“Let’s just throw AI spaghetti at the wall and see what sticks, is not a strategy. Incorporating AI into your architecture is a strategy, and that’s where you develop your governance and your spend.”
Richard Patterson
Chief Data and AI Officer, Bureau of the Comptroller and Global Financial Services, U.S. Department of State
This strategic view is essential because orchestration can extend across many modernization priorities at once. A framework developed for compliance evidence gathering may also support ATO workflows. A data cataloging workflow may improve pipeline governance. A process for routing AI-generated outputs for human review may become the basis for broader responsible AI controls.
The opportunity is to create reusable patterns for modernization. Instead of treating each AI use case as a separate project, agencies can build common orchestration capabilities that support multiple functions. Over time, this can help reduce duplication, improve consistency, and make governance easier to enforce.
Steven Walters, CEO of Property Management Resources LLC, frames this as a shift to platform-oriented approach.
“Organizations avoid rebuilding governance for every use case by treating AI as a platform capability rather than a series of one-off projects. Audit trails, permission checks, escalation points, and human review become standard building blocks that any new workflow inherits by default, so only the mission-specific logic changes from case to case.”
Steven Walters
CEO, Property Management Resources LLC
The clearest sign of progress is not the number of pilots launched, or AI tools procured, but whether those capabilities are becoming part of everyday mission delivery. It also changes how leaders should measure the maturity of AI integration.
“We’ll know we’ve moved beyond the hype when we stop talking about AI initiatives and simply talk about mission outcomes.”
Richard Patterson
Chief Data and AI Officer, Bureau of the Comptroller and Global Financial Services, U.S. Department of State
That is the long-term promise of workflow orchestration in the public sector. Making AI part of the operating fabric of government in support of faster, safer, and more accountable modernization.
_
This article is an excerpt from Distributed AI for the Mission. Download the full report to explore how public sector organizations can turn AI pilots into governed workflows and deploy AI infrastructure closer to where missions operate.
