AI Agents for Medical Practices: 83% Use AI, Only 26% Changed a Single Job

Ask a medical practice administrator whether AI has arrived and the answer is almost certainly yes. Ask whether it has changed anybody's job, and the answer gets a lot quieter.

That gap shows up clearly in the weekly MGMA Stat polls the Medical Group Management Association runs with medical group leaders. Read together, the 2026 polls describe an industry where AI is now nearly universal in the exam room, still patchy in the front office, and mostly ungoverned everywhere in between.

AI is in 83% of practices now

An August 4, 2026 MGMA Stat poll found that AI plays some role in patient visits at 83% of medical groups, up from 71% when MGMA asked the same question a year earlier. The poll had 189 applicable responses.

The more interesting change is depth, not reach. Among practices using AI at all, about one in five now use it in at least three-quarters of their visits, up from about one in ten the year before. MGMA's read is that existing users expanded AI into more visit types rather than a wave of first-timers driving the growth.

But when MGMA asked what AI had done to staff workload, the answers were not the clean "it saves time" story vendors tell. Across the coded replies, 37% described work shifting to other tasks, while 29% described an actual reduction. AI is moving work around more often than it is removing it.

Most practices haven't changed a single role

A June 2, 2026 MGMA Stat poll asked whether practices had redesigned a role or adjusted staffing with the help of AI in the past year. 68% said no. Only 26% said yes, and 5% were unsure (260 applicable responses).

The practices that did make changes described them as "small and practical," and almost all of it sat in the same place: the front office, the revenue cycle and the call center. Respondents talked about automating scheduling, registration, prior authorization, billing and routine correspondence. Some left positions unfilled, some put AI agents on the phones, and some stopped replacing scribes who left. Clinical changes were narrower.

The 68% majority were mostly "watching and waiting," pointing at the same use cases (revenue cycle, scheduling, prior auth, call handling) with vague timelines of three to twelve months.

The front office is where the first pass is moving

MGMA's June analysis also recaps a February 10, 2026 poll on front-office AI priorities. Practice leaders ranked scheduling (31%), calls (27%), registration and eligibility (23%) and prior authorization (16%) as their top operational targets.

The pressure behind that list is not subtle. The same analysis cites MGMA's own regulatory burden research, in which 92% of practices said they had hired or reassigned staff solely to handle prior authorization volume, and the CAQH Index finding that providers spend an average of 11 minutes on each electronic prior authorization and 16 minutes on each one done through a payer portal. That time scales directly with patient volume unless the workflow itself changes.

This is exactly the shape of work an AI agent handles well: a high-volume queue where the first pass is repetitive (check eligibility, pull the right documents, fill the payer's form, chase the status) and the exceptions need a person. MGMA describes the emerging pattern the same way: software takes the first pass, and humans take the cases AI flags as low-confidence, complex or patient-facing.

Practices are already building their own tools

One finding should get the attention of any practice leader who assumes AI only arrives through the EHR vendor. A September 22, 2026 MGMA Stat poll asked whether practices had built their own app or workflow tool in-house using AI. 34% said yes, another 8% had tried but not launched, and 10% were unsure (228 applicable responses).

That 10% "unsure" matters. MGMA points out that leaders may not have full visibility into tools being built inside their own organizations. The projects described ranged from dashboards and policy documents to charting support and patient-facing tools, and MGMA's advice is blunt: "we built an app" is not a risk category. What the tool can see, change or generate is what matters.

Governance is running behind adoption

A January 20, 2026 MGMA Stat poll found that only 20% of medical groups have AI governance or a formal AI-use policy, with another 22% working on one. 56% said no (328 applicable responses). That is progress from MGMA's 2024 research with Humana, when 73% of organizations had no formal governance structure for AI, but it still means most practices are running AI tools without written rules about what data they can touch or who reviews their output.

MGMA's governance guidance is practical: an executive owner, a small working group, a lightweight intake process for any new tool, plain-language data rules, and a defined meaning for "human review" rather than a vague "human in the loop."

What this means for practice leaders going forward

The MGMA data points to a fairly clear playbook for the next year:

• Start in the queues, not the exam room. Scheduling, inbound calls, eligibility checks and prior authorization are where leaders already want help, and where an agent's first pass is easy to check.
• Pair every automated workflow with a named human review point. MGMA recommends exactly this, along with tracking the metrics the automation is supposed to move: claim turnaround, days in A/R, note completion time, no-show rate.
• Expect shifted work, not vanished work. The 37% vs. 29% split on workload is a useful reality check. Budget for the people who will handle the exceptions and review the agent's output.
• Write the policy before the tool count grows. With a third of practices already building their own tools, the intake process matters more than any single vendor decision.

AI is already in the room at most practices. The next step is getting it into the work in a way someone can measure, review and defend.