Epic Agent Factory [What It Changes]

TL;DR

• Epic Agent Factory is Epic's no-code platform for health systems to build, run, and monitor their own AI agents, previewed at the 2026 Users Group Meeting, with a wide release planned for 2027. That date is an availability boundary. This hub is the preview, not a generally available SKU.

• Forbes contributor David Chou (August 25, 2026) places it next to Penny, Cogito Assistant, and about 120 AI features organizations can adapt. Fierce Healthcare independently reported the same 120-feature, 2027-wide-release pair from UGM.

• In-chart Coverage Requirements Discovery and real-time prior authorization with payers such as UnitedHealthcare and Aetna are adjacent UGM themes. They are not Agent Factory.

• A two-truck HVAC shop, a 10-person agency, or a solo clinic will not buy Agent Factory. They should care because the hospital they send work to will start shipping agents that change referrals, prior-auth questions, and scheduling — felt as a portal change, not a software purchase.

Key Takeaways

• Wide release in 2027 means you cannot treat Agent Factory as a SKU you order this week for a community practice.

• Attendance figures from UGM disagree across outlets. Report them separately. Do not average them.

• Curiosity (Cosmos-trained predictions) and Ergo (visit workflow) are different products with different dates.

• The first honest test, if you are even an early adopter, is one agent that drafts, cites the chart, and stops. Not one that signs an order.

The answer in plain English

Epic Agent Factory is a no-code shop, previewed at UGM 2026, where a health system can take one of Epic's available AI features, shape it, or make an agent itself — with wide release planned for 2027. As of August 25, 2026, that is Chou's reporting from Verona. Fierce Healthcare's UGM write-up, dated August 19, 2026, uses Judy Faulkner's line: "You can take it, shape it, make it."

A two-truck HVAC company will not log into Hyperspace. A 10-person marketing agency will not build a Cosmos model. A solo clinic on a different ambulatory EHR will not get Agent Factory because a hospital down the road turned it on. They should still watch the pattern. When a system of record lets the customer build agents that write back into that record, every smaller shop that touches that record — referrals, prior auth, scheduling, records requests — inherits the agent's mistakes. The clinic's version of this is already familiar: a hospital portal that suddenly wants a different packet, or a payer that now checks coverage in the order screen. Agent Factory is that pattern with a 2027 door.

This page is current as of August 25, 2026. It is not a GA data sheet.

Who should use this page

This page is for a CIO, CMIO, revenue-cycle lead, or ambulatory owner who heard "Epic is building agents" and needs the preview separated from Ergo, Curiosity, Penny, and prior-auth checks.

Red flags: budgeting 2026 production on a 2027 wide release; treating 120 features as 120 validated outcomes; letting an agent send an order.

What UGM actually said

According to Forbes, over 20,000 health system leaders gathered in Verona and more than 60,000 joined the General Session online. Chou, a Forbes contributor, frames Agent Factory as a no-code platform for health systems to create their own AI agents, with a wide release planned for 2027. Wide release is planned for 2027. That is Chou's reporting, not an Epic contract.

According to Forbes, Penny handles conversational actions, Cogito Assistant produces dynamic visualizations, and Agent Factory gives organizations access to roughly 120 AI features they can adapt. Chou notes that nobody knows yet what it costs to build inside Agent Factory — developer time, staff hours, token spend — or the ROI. August 2025 is when Chou says Epic launched Emmie (patient chatbot), Art (clinician AI), and Penny (administrative AI). This year's UGM added Ergo as an assistant that combines those tools. Ambient AI, in Chou's July 2025 Forbes piece, is already becoming a commodity. Abridge's same-week announcement is Chou's example of a point-solution response. Reuters is Chou's cite on Silver Lake / Workday talks as ERP context for Epic's 2029 suite plans.

According to Fierce Healthcare, more than 8,400 people attended in person and an estimated 70,000 attended virtually and in-person, per an Epic spokesperson. 8,400 in-person is Fierce's UGM figure. It does not match Chou's 20,000. Do not blend them. Fierce also reports Epic at 43.7% of the acute-care EHR market, software in more than 3,700 hospitals, and medical records for more than 325 million patients.

According to Fierce Healthcare, Faulkner tracked 167 projects: 84 done on time, 7 completed or soon, 76 in progress and on track. Agent Factory is the platform for building, running, and monitoring AI agents. Early adopters are using it today, Epic said, and it will be widely available in 2027. ECU Health built an agent to summarize transfer-center requests that "saved an estimated 20 hours a week," Epic told Fierce — vendor-reported. Cosmos holds over 320 million patients and 23 billion encounters. Curiosity, the generative model on that data, is planned as an EHR component in March 2027, with 20 organizations validating it. Chart with Art is live in more than 70 specialties; summaries are in use at over 300 organizations; 68% of surveyed users said visit summaries that include Care Everywhere notes highlighted new information. In March, Epic said more than 85% of customers use Epic AI. Chart with Art for bedside nursing is live at 11 health systems.

According to Becker's Hospital Review, the August 13 preview said Epic would host more than 20,000 attendees in person, with the General Session drawing over 60,000 total viewers on Tuesday, August 18, and nearly 500 presentations. Charité would join as the first Epic organization in Germany. Becker's also noted 914 million Care Everywhere records exchanged in June. That preview is dated before the meeting; Fierce is dated after. Becker's Faulkner session note, what's coming in Epic, and the Honor Roll are linked from that preview.

Healthcare AI Guy's August 25 weekly restates Agent Factory with 120 prebuilt AI capabilities plus tools for health systems to build their own agents, Ergo's visit flow, Chart with Art across 70+ specialties and summaries at 300+ organizations, and Curiosity on Cosmos data from 320M patients and 23B encounters. That newsletter points to the Fierce UGM story as its link. Use Fierce for those counts when they overlap.

Sources: Forbes; Becker's; Fierce Healthcare.

Adjacent UGM products that are not Agent Factory

Coverage Requirements Discovery and real-time prior authorization are Chou's third theme and Fierce's interoperability block. Four health systems, including Ochsner and Denver Health in Fierce's telling, went live with real-time prior-auth checks. Epic collaborated with UnitedHealthcare, Network Health, and Aetna, using the Coverage Requirements Discovery API, and is testing with 16 additional payers. Fierce's prior-auth story is the companion. Selected F coverage already owns the in-chart prior-auth check. This hub does not.

Ergo Visit is live at Ochsner and five Garden Plot physicians, per Fierce. Becker's Ochsner Ergo item is the first-to-use claim. Art, Emmie, and Penny remain named copilots. Ask Emmie at Ochsner saved more than 1,000 hours for call-center staff over 10 months, Fierce reports as Ochsner's figure. Community Health Network reported 73% fewer billing-related InBasket messages.

Curiosity is March 2027, not Agent Factory. Yale's medRxiv evaluation on 3,000 patients is Fierce's research aside; it is a preliminary, non-peer-reviewed article. Cosmos look-alike was used 25,000 times in the last year. Best Care Choices is live at nearly 60 organizations. Care Everywhere exchanged 9 billion records last year; 2,150 hospitals and 58,000 clinics on Epic are connected to TEFCA, Fierce reports.

EpicOps is the ERP thread. According to Epic, Teamwork is the first EpicOps application. Parkview's CMIO said it takes 75% less time to build provider schedules. The post is dated June 18, 2026. Dubai Health is the first organization to implement Teamwork for nursing and the first health system outside the US to adopt EpicOps. Chou says University of Iowa Health has started using EpicOps for provider scheduling, with cost accounting in May 2027. Fierce says credentialing and cost accounting modules are expected mid-2027, supply chain and financials at the end of next year, then HR and payroll. Those dates are ERP, not Agent Factory.

Healthcare AI Guy also flags an FDA discussion paper on generative AI devices and a JAMA Perspective. The FDA paper is labeled discussion only, not guidance. It defines agentic AI systems as GenAI-enabled systems that autonomously plan and execute multi-step tasks, use external tools, or take actions across a sequence of steps, and it asks how elevated risk from autonomous multi-step action should be reflected in oversight. The AI-enabled medical devices list and the November 2024 DHAC executive summary are the FDA pages that paper cites. Agent Factory is a health-system build shop inside an EHR. Whether a given agent is a device is a regulatory question this hub cannot close. The JAMA Perspective is HAIG's separate "doctor + AI" debate, not an Epic SKU.

A clinic comparing Epic versus athenahealth is in the ambulatory market Agent Factory is not selling to this year. The state of healthcare automation and the small medical practice automation guide are the right next reads for bounded queues that do not wait on 2027.

USTA analysis: 84 of 167 on time

Working only from Fierce's Faulkner project tracker: 84 of 167 projects were done on time. That is 84 ÷ 167 ≈ 50.3% done on time by that count, with 7 more completed or soon and 76 in progress. Inputs: 84, 167, 7, and 76 in Fierce Healthcare. Faulkner used the tracker to argue "what you hear here is very likely to happen." The arithmetic is a completion mix, not a guarantee that Agent Factory ships in 2027 as previewed. 50.3% of tracked projects were done on time.

Source for the four counts: Fierce Healthcare. 50.3% is 84 ÷ 167.

Why now: the bolt-on squeeze

Chou's logic is the storefront becoming a build shop: keep health systems building inside Agent Factory instead of buying outside agents, the way Salesforce and Workday built platform gravity. Fierce's Seth Howard line is that Art, Emmie, Penny, Agent Factory, Curiosity, and the network are "a single integrated system." The constraint that broke, on that telling, is "the EHR was a storefront for other people's apps." The new constraint is "the EHR wants to be the place you manufacture agents." Cost and ROI are, Chou says, unknown.

Teams already routing prior-auth packets, referral documents, or exception lists through US Tech Automations should not rebuild those queues inside a 2027 preview. Keep the reviewer, the source document, and the stop button. If a health-system partner later exposes an Agent Factory action, treat it as another model behind the same step — a swap, not a rebuild.

What this does not establish

It does not establish that Agent Factory is generally available. It does not establish a token price. It does not establish that 120 features are 120 safe autonomous actions. It does not establish FDA clearance for a customer-built agent. It does not reconcile 20,000 versus 8,400 in-person attendees.

Midsize solutions and data-extraction agents are USTA routes for shops that need a cited draft today. They are not Epic.

A buyer's evaluation sequence

US Tech Automations can keep the exception record if a health-system agent returns a packet the clinic must review. It should not be configured to sign the order.

A week-one test that still cites the chart

If you are not an early adopter, the week-one test is simpler: do not budget Agent Factory as a 2026 GA SKU. Write 2027 on the roadmap. Keep today's extract-and-review queues. If a hospital partner starts sending packets that look agent-shaped, treat them as inbound documents with a reviewer, not as signed orders.

If you are an early adopter, pick one draft-only task. Fierce's ECU Health example is a transfer-center summary. That is the right shape: a handoff document a person still reads. Do not start with an agent that enters an order, closes a prior-auth, or messages a patient. Chou said cost and ROI inside the factory are unknown. Believe that. Time the staff hours and the token bill on a known-answer pack of real transfers before you shape a second feature from the 120.

Attendance figures will show up in vendor slides. Do not average 20,000 and 8,400. Cite the outlet. The disagreement is itself a reminder that UGM reporting is journalism, not a contract. Wide release in 2027 is the availability line that both Chou and Fierce carried. Curiosity in March 2027 is a different line. Ergo Visit at Ochsner is a different product. Prior-auth CRD with UnitedHealthcare and Aetna is a different product. A slide that stacks them as "Epic AI" is a shopping error.

Keep a written stop: one uncited action, or one draft that a physician cannot match to the chart, and the agent pauses. The FDA discussion paper is not guidance, but it names the risk of agentic systems that take multi-step actions with less chance of human review. Your stop rule is how a health system answers that question without waiting for a final FDA document.

A community clinic will not run this test. It will notice when the hospital portal wants a different packet. The small medical practice automation guide is the clinic's version: one inbound document, one proposed field, one named reviewer. Do not wait for Agent Factory to become the place you manufacture that queue. You can already own it.

If Epic later publishes a customer-built-agent outcome table, read the denominator. Until then, the numbers that belong in a board deck are 120 features, 2027 wide release, 84 of 167 projects done on time on Faulkner's tracker, and your own transfer tally — or, if you are not an early adopter, a written decision to wait.

Signal vs Speculation

Demonstrated signal: as of UGM week 2026, Epic previewed Agent Factory as a no-code agent platform with about 120 features to adapt and a wide release in 2027. Chou (August 25) and Fierce (August 19) both carry those two facts. Early adopters are using it, Epic said; ECU's 20-hour claim is Epic-reported. Attendance counts conflict across Forbes/Becker's and Fierce. Curiosity is March 2027. Prior-auth CRD and Ergo Visit are live in named systems. FDA's generative-AI discussion paper is not guidance. Teamwork's 75% schedule-build claim is a June 18 Epic customer quote about EpicOps, not Agent Factory.

Our read: over 12–36 months, large Epic shops will prototype one draft-only agent in the factory and leave bolt-on ambient and prior-auth tools in place until contracts renew. Some will try to replace a niche vendor because "good enough and native" is cheaper than a second security review. Community clinics will feel this as portal and packet changes, not as a product they buy. Empty agents and unreviewed writes still do not count. If 2027 slips, the preview was still a real UGM event; it was never GA.

Frequently asked questions

What is Epic Agent Factory?

It is Epic's no-code platform, previewed at UGM 2026, for health systems to take, shape, or make AI agents that run in Epic workflows, with wide release planned for 2027.

Is it generally available?

No. Epic said early adopters are using it and that it will be widely available in 2027. This hub is the preview.

How is it different from Penny or Ergo?

Penny is an administrative copilot. Ergo is the visit workflow that combines Art, Emmie, and Penny. Agent Factory is the shop for building and monitoring agents, including shaping existing features.

Why do attendance numbers disagree?

Forbes and Becker's used a 20,000 in-person / 60,000 session figure. Fierce reported 8,400 in-person and about 70,000 virtual and in-person after the meeting. This page reports both.

Does this include real-time prior authorization?

No. Prior-auth checks and Coverage Requirements Discovery are separate UGM items. Do not fold them into Agent Factory.

Can a solo clinic buy this?

Not as a 2026 SKU. A clinic may still have to answer packets and portal prompts that a health-system agent generated.

How should a team plug a 2027 agent preview into work they already run?

Do not. Keep today's extract-and-review queues. If an early-adopter partner exposes one draft-only agent, treat it as a model behind a step you already own, with the same reviewer.

If you already route documents, exceptions, and approvals through automated steps, map an agentic workflow that stays draft-only until a person signs. US Tech Automations is the homepage for that routing layer, not an EHR.