Epic’s Agent Factory is drawing attention for a practical reason: it brings AI-enabled automation closer to where healthcare work already happens. For health systems, that is a meaningful shift. Instead of adding another layer of AI tooling, organizations have an opportunity to improve the daily operational work already within Epic, including:

• Referrals.
• Scheduling.
• Documentation support.
• Revenue cycle activities.
• Patient messaging.
• Other high-volume workflows.

That means leadership’s first question should not be, “How quickly can we turn this on?” but instead, “Which workflows are ready to be automated, and which ones need to be fixed first?”

Agent Factory will not repair a broken process simply because it is embedded in Epic, but it will accelerate the given process. When the workflow is standardized, governed, and well understood, an agent may reduce burden, improve consistency, and support more reliable execution. When the workflow is fragmented, informal, or dependent on local workarounds, the same agent may scale those weaknesses faster than leaders can manage them.

What Agent Factory Actually Is

At its simplest, Agent Factory gives Epic customers a way to build AI agents inside the Epic environment. Teams can assemble agents through visual steps that look more like workflow design than traditional software development. Those agents can gather information, evaluate a situation, and support action in the same system that clinicians and operations staff are already using. By operating within Epic, agents can be designed closer to the point of work, which can:

• Reduce integration friction.
• Improve access to workflow context.
• Shorten the path from idea to operational use.

The potential is real, but it should be viewed with discipline. Two distinctions are worth making.

• First, using AI inside Epic is not the same as being ready to deploy agents that act within workflows. Many current AI tools draft, summarize, suggest, or flag. Those capabilities can be useful, but they differ from those of agents that initiate, route, complete, or escalate work. The more a tool moves from suggestion to action, the more governance, accountability, and workflow discipline matter.
• Second, Agent Factory is still maturing. Early adoption should be treated as a learning agenda, not a race to broad deployment. For most health systems, the near-term opportunity will be to use agents to support people, reduce avoidable work, and improve consistency—not to remove human judgment from operational decision-making.

What the Numbers Tell Us

Epic’s footprint gives Agent Factory a natural path to scale. Because Epic is already embedded in the daily work of many health systems, agent-enabled automation will likely move quickly from innovation discussions to operational planning. Leaders will be asked where agents can reduce burden, improve throughput, support access, strengthen revenue cycle performance, or make patient communication more consistent.

There is a governance gap. Many organizations still struggle to explain what an AI tool did, when it did it, why it did it, and who intervened. That may be acceptable when a tool only drafts or summarizes. It becomes more serious when an agent begins to influence or complete operational work.

Why This Matters for Operations Leaders

The core management lesson is not new. Automation makes strong operations stronger, but it can also make weak operations more visible and more expensive. Health systems have seen this before: when technology is applied to a flawed workflow without redesign, it often preserves the problem in a more digital form.

Agent Factory raises the stakes by making it easier to build automation within Epic. That ease of use is part of its value. It is also what makes governance so important. In the past, adding a third-party tool often forced teams to map the process, define handoffs, clarify exceptions, and assign accountability before launch. With Agent Factory, the tool may be easier to build than the workflow is to govern.

Three risks deserve particular attention.

1. Messy Processes Will Scale

   An agent pointing at a fragmented workflow will not make it clean. It will speed up the fragmentation. If referral management varies by department, medication reconciliation relies on informal workarounds, or prior authorization steps differ by team, the agent may reproduce that variation at scale.

2. Variation Will Become Harder to See

   Healthcare operations depend on knowledge that is often learned through experience rather than captured in policy. Staff know which clinics handle intake differently, which providers prefer referrals routed a certain way, which payers require extra documentation, and which exceptions truly matter. An agent will not know that unless it is made explicit.

3. Cost Tends to Track Usage

   For operations leaders, agents should be managed as both technology capabilities and recurring operating costs. Each review, recommendation, draft, routing decision, or completed task adds cost. In a pilot, that may be manageable. At enterprise scale, a poorly designed agent moving through thousands of records can become an expensive way to automate inefficiency.

What Leaders Should Do First

The right conclusion is not to slow down indefinitely but to sequence the work correctly. The most important preparation for Agent Factory happens before an agent is built: in workflow design, governance, measurement, and workforce readiness.

Before scaling agents in Epic, leaders should focus on five practical moves.

1. Standardize Workflows Before You Automate

   Start with two or three workflows that matter, but are still manageable. Good candidates may include referral routing, appointment preparation, follow-up documentation, claims support, or patient message triage.

   Before building an agent, document how the work actually happens today.

   • Identify variation by site, role, specialty, and exception type.
   • Remove unnecessary handoffs.
   • Simplify the process.
   • Agree on the standard work the agent will support.

2. Create Real Oversight

   Health systems need an oversight model with real authority. The group should be able to approve, pause, modify, or retire agents and should include representation from operations, clinical leadership, IT, compliance, privacy, legal, security, and the front line.

   Each agent should have a named owner, a clear purpose, a defined scope, a risk rating, and a monitoring plan. Leaders should maintain an inventory of agents and document when people intervene, override, or correct agent actions. If something goes wrong, the organization should be able to reconstruct what happened, why it happened, who was affected, and how the issue was addressed.

3. Start Small and Keep It Reversible

   Initial agents should support low-risk, standardized, high-volume work. They should assist users before they act independently, and human review should remain in place for exceptions or when uncertainty arises. Accuracy is not enough. Leaders should also track safety, staff burden, rework, cycle time, patient experience, cost, and escalation rates. An agent that looks accurate but creates more downstream work is not successful.

4. Manage Costs from Day One

   Because agents create recurring usage-based costs, financial visibility should be built in from the start. Track cost by agent, workflow, volume, and outcome. Compare that cost with simpler alternatives, such as fixing the workflow, using rule-based automation, improving reporting, adjusting staffing, or providing targeted training. Agent Factory should not become the default answer to every operational problem.

5. Invest in People

   Agent Factory will make the Epic and operations teams more capable. It will not make up for unclear ownership or under-resourced support teams.

   Health systems should invest in people who understand both Epic and the operational workflows that agents will affect. Stakeholders need enough AI literacy to ask practical questions, recognize risk early, and guide safe adoption.

---

---