A quality coordinator opens her worklist on a Monday and sees hundreds of open care gaps. Screenings due, labs missing, medications lapsing, follow-ups overdue. Each one, handled the traditional way, is a small project: read the history, find the right system, make a call, wait, call again, write it up. There are not enough Mondays in the quarter to clear the list by hand. This is the actual problem in value-based care operations, and it is the problem an execution agent is built to take off her desk.
Below is the anatomy of a single gap, from the moment it appears to the moment it is closed and documented. I will use a medication-adherence gap as the worked example, because it touches everything: members, pharmacies, prescribers, claims, and the Stars measures that depend on all of them.
Step 1: Pull the context
Before doing anything, the agent assembles what a careful coordinator would gather first, except it does it in seconds instead of across four tabs. From one canonical record per member, it reads the relevant history: the previous call notes (was this member reached last month, and what did they say?), claims data (when was the last fill, what does the days-supply imply?), pharmacy and lab signals, and any open items from other teams.
This matters because context determines the right move. A member who already explained a copay problem needs a different action than one who simply forgot. Without the full record, the agent would repeat the same call the member already ignored. With it, the first action is informed.
Step 2: Prioritize across data sources
The agent does not work the list top to bottom. It prioritizes across the same signals a good manager would weigh: how close the member is to falling below the adherence threshold, how many days remain in the measurement period, whether the gap also affects a triple-weighted measure, and whether the same member has other open items that should be handled in one touch rather than three.
The goal is to avoid the failure mode every operations leader knows: three teams calling the same member in the same week about three different things, because each was working its own list off its own tool.
Step 3: Decide who owns it
This is the step that separates a real agent from a script. For each gap, the agent triages: is this a low-risk, well-bounded case it can own end to end, or one that should route to a specific human?
- Agent-owned. A routine refill follow-up where the member is reachable, the regimen is stable, and the next action is clear. The agent takes it.
- Routed to a human. Anything that needs clinical judgment, a sensitive conversation, or a decision outside the bounds it was given. The agent routes it to the right coordinator, with context attached, rather than guessing.
The decision is bounded by rules the operations team sets. The agent does not invent its own scope. It works the cases it was cleared to work and hands up the rest.
Step 4: Take the action (the medication-adherence example)
For an agent-owned medication-adherence gap, the sequence runs like this, without anyone on the team touching it:
- Check with the pharmacy. A voice agent calls the pharmacy to confirm fill status. Was the prescription picked up? Is there a refill on file? Is it stuck on a prior authorization?
- Reach the prescriber. If the issue is an expired or missing prescription, the agent leaves a structured voicemail for the prescribing physician, stating the member, the medication, and the specific action needed.
- Wait, then follow up automatically. The agent sets its own follow-up. When there is no reply within the window, it calls back, rather than waiting for a human to notice the silence.
- Work the portal if needed. If a prior authorization is the blocker, a computer-use agent logs into the payer portal and submits or checks the authorization the way a person would.
- Escalate only when needed. If the case stalls in a way that needs judgment, it goes to a live coordinator with the full thread attached.
By the time the team looks at this member again, the authorization is resolved and the medication is filled. The coordinator did not spend her Monday on it. She spent it on the cases that genuinely needed her. We make the broader version of this argument in AI agents vs. analytics dashboards: the leverage is removing the work that should not have required a human.
Step 5: Document everything
Every step writes back to the canonical record: who was called, what was said, what was submitted, when the follow-up fired, and why the case was escalated if it was. This is not a nicety. It is what makes the work auditable and what lets the next person, or the next agent, pick up current state instead of a stale snapshot. Under SOC 2 Type II and HIPAA, with full audit trails, the record of every action is the product, not a byproduct.
Voice agents and computer-use agents, briefly
Two capabilities make the sequence above possible, and they are worth defining plainly.
Voice agents
A voice agent places and receives phone calls on behalf of the team. It can call a member about a refill, call a pharmacy to confirm fill status, or leave a structured voicemail for a prescriber. Critically, it captures what was said and writes the outcome back to the record, so a call becomes data instead of a note someone may or may not transcribe.
Computer-use agents
A computer-use agent operates software the way a person does. It logs into a payer portal or EHR, navigates the screens, and enters or extracts data, including in systems that expose no API. In value-based care, that is most of them: critical data still lives behind portals and SFTP drops that were never designed to integrate. Rather than wait for an integration that may never ship, the agent uses the interface a human already uses.
Where this lands: HEDIS and PDC
This is not an abstract automation exercise. Many care gaps are quality and adherence measures with direct financial weight. A medication-adherence gap maps to proportion of days covered (PDC), which feeds the triple-weighted Part D adherence measures that move Star Ratings. A screening gap maps to a HEDIS measure. We walk through why a few points of PDC separate a 4-star contract from a 2-star one in our PDC math guide.
The point of closing the loop, rather than just reporting it, is timing. A gap closed in May counts. The same gap "identified" in May and still open at the quarterly reconciliation is a loss. The agent's job is to make the close happen while it still matters.
The measure of an execution agent is simple: when your team next looks at the member, is the work done, or is it still on the list?