A CureMD practice should rely on built-in eRx refill tools when volume is low, providers are few, and protocols are simple, and should add an AI refill agent when several providers, high request volume, or inconsistent protocols make the manual lookup work the bottleneck. Native tools transmit and track renewals; an AI refill agent adds the protocol checking, chart review, and pre-drafted renewals on top. The decision turns on how much of your refill time goes to investigating requests rather than signing them.
What do built-in eRx refill tools actually do?
Built-in electronic prescribing tools handle the transmission side of refills. They receive renewal requests from pharmacies, show them in a queue, let a provider or delegate approve, deny, or change them, and send the response back electronically with the right audit trail.
That is real, necessary functionality, and for controlled substances it is mandatory. Federal rules at 21 CFR 1311.120 set requirements for the application, including two-factor authentication for signing and an audit trail of creation, signing, and transmission. Your EHR's native eRx module is built to meet that standard, and you should keep it as the system that actually sends prescriptions.
What native tools generally do not do is the thinking that happens before the click. A nurse or MA still opens each request, finds the patient's last visit, checks whether labs are current, compares the medication to the protocol, and decides which lane the request belongs in. The tool gives you a queue. The investigation is still manual.
Feature sets differ by CureMD version and configuration, so ask your account team exactly what your current setup includes before you decide anything. The comparison below assumes a typical native refill queue and not any specific edition.
What does an AI refill agent add on top?
An AI refill agent automates the investigation. It reads each incoming request, matches it to the patient and the active medication, runs your protocol checks against chart data, and either drafts a renewal for sign-off or routes the request to a person with the reason attached.
It does not replace eRx transmission. The prepared renewal still goes through your EHR's prescribing workflow, and a licensed provider still approves it. The agent sits in front of that step and does the lookups so the provider confirms instead of investigates.
Honey Health's refill agent is one example of this layer. It works inside the practice's existing EHR workflow and leaves the approval step with the provider. The general pattern, whichever vendor you choose, is the same: native tool for sending, added layer for preparing.
How do the two approaches compare?
The clearest way to compare them is on the dimensions that affect your daily workload and risk.
Work per request. With native tools alone, each request needs a manual chart review. With an added agent, the review is done before the request reaches a person, so routine renewals take one confirmation.
Protocol enforcement. Native tools do not apply your protocol; staff do. That works when one person handles every request and breaks down when several staff members apply the rules differently. An agent applies the same rules in the same order every time.
Auditability. Native tools log what was sent and who signed it. A good agent also logs which checks ran and why a request was routed where it was, which helps when you review decisions or train new staff.
Setup effort. Native tools are already there. An agent requires you to write down your protocol, connect to the chart, and run a pilot, which takes a few weeks.
Provider time. Providers touch fewer requests under an agent, since routine ones arrive prepared and exceptions arrive with context. The saving is real only if your volume is large enough to matter.
Cost model. Native tools are usually bundled in what you already pay. An agent adds a cost, so the question is whether the staff time it returns outweighs it.
When are native tools enough?
Native tools are enough more often than vendors admit. If your practice has one or two providers, a low and steady refill volume, a short list of medications, and a staff member who knows the protocol by heart, adding another layer buys little.
Native tools are also enough when your backlog is caused by something else. A practice whose refill requests wait because nobody owns the queue has an ownership problem, not a software problem. Assigning a daily owner and a turnaround target will do more than any tool.
Here is a quick test. If your nurses and MAs spend less than a few hours a week on refill lookups, and turnaround is within your target, keep what you have. Revisit when volume grows or a second site opens.
When does an added layer pay off?
An added layer starts to pay off when the investigation work, not the signing, consumes the time. Four situations usually push practices over the line.
- Several providers with different habits. When each provider has their own preferences, staff spend time remembering who wants what. A shared, enforced protocol removes that overhead.
- High volume. In one health system study, refill requests made up 178,160 of 472,411 out-of-visit tasks, and providers without scribe support took about three hours on average to close each one. A large group sees that pattern at scale.
- Inconsistent turnaround. If the median looks fine but the 90th percentile is days long, requests are getting stuck in an unowned queue.
- Multiple locations. A group with several sites wants the same rules everywhere, which is hard to maintain by training alone.
The scale of possible savings is illustrated by a Geisinger rheumatology group that moved routine refills to a pharmacist-led protocol. Physicians went from handling all refills to about 26%, and the group reported roughly 41.5 hours a month returned. That was a human-run protocol, but it shows how much of refill work is protocol-driven and therefore automatable.
What about safety and control?
Safety is the right question to ask about any added layer, and the answer should be specific. An agent should never sign a prescription, and it should never approve a controlled substance. Those decisions belong to a licensed prescriber working inside your EHR's certified prescribing function.
Ask any vendor four direct questions. Which request types does it route to a person automatically? Can you see which protocol checks ran on every request? Can you change the rules without a support ticket? What happens when chart data is missing or ambiguous? A tool that treats missing data as a reason to escalate is safer than one that guesses.
Start in a review-everything mode. For the first couple of weeks, staff compare what the agent would have done to what they actually did. Disagreements show you whether the protocol or the integration needs fixing before the agent handles real volume.
What hidden costs should you weigh on each side?
Both options carry costs that do not show up on an invoice. Native tools alone cost staff time, and that time compounds as volume grows. If your nurses spend two minutes on a lookup, a modest 400 requests a week means more than 13 hours of lookups before anyone has made a clinical decision. Check that arithmetic with your own volume and your own touch time.
An added layer costs attention. Someone has to own the protocol, review the exceptions, and keep rules current when a provider changes their mind or a payer changes a formulary. Practices that treat the agent as set-and-forget find that decisions drift away from policy. Budget a monthly review of routed requests, and name the person responsible.
There is also a switching cost to consider in both directions. Moving from manual to automated means retraining staff on a new workflow. Staying manual means every new hire has to learn the unwritten rules. Neither cost is large, but both belong in the comparison.
What would a 30-day pilot look like?
A pilot gives you evidence instead of vendor promises. Pick one medication class with high volume and low risk, such as stable maintenance medications for patients seen within the past year, and run the agent in review mode for the first two weeks.
In weeks one and two, staff handle requests as usual while the agent produces its own recommendation for each. You compare the two and log every disagreement. In week three, switch the agent's prepared renewals into the live queue for provider confirmation, still on that single class. In week four, measure turnaround, touch time, and the share of prepared renewals providers changed.
Set a decision rule before you start. For example: expand if prepared renewals are accepted without changes at a rate you consider safe and touch time drops meaningfully; fix the protocol or integration if disagreements cluster around one cause; stop if the agent cannot handle your chart data. A pilot with a stated rule ends in a decision rather than a debate.
How should you decide?
Use your own numbers. Measure weekly refill volume, median and 90th percentile turnaround, and the minutes staff spend per request. Then estimate the weekly hours spent on lookups rather than decisions.
If those hours are small, keep native tools and fix ownership. If they are large, or if inconsistent protocols are causing errors, pilot an agent on one medication class and compare results. Keep the native eRx module either way, because it is the system that sends prescriptions.
The choice is not either-or. It is a question of whether the layer in front of your queue is a person doing lookups or software doing them while people do the deciding.
Frequently Asked Questions
Does an AI refill agent replace CureMD's eRx module?
No. The eRx module remains the system that transmits prescriptions and meets controlled-substance requirements. An agent prepares and routes requests in front of it, and providers still approve renewals through the normal prescribing workflow.
Is an AI refill agent safe for controlled substances?
An agent should not approve or sign them. Federal rules require the prescribing practitioner to sign electronic controlled-substance prescriptions with two-factor authentication. A well-designed agent routes those requests to the prescriber and may gather chart context to speed the review.
How do we know if our practice is big enough to benefit?
Look at hours, not provider count. If staff spend several hours a week on refill lookups, or if turnaround is inconsistent, an added layer likely helps. If refill work is a small, steady task, native tools are probably enough.
How long does it take to add an agent to a CureMD workflow?
Plan on a few weeks, most of it spent writing your protocol and running a review pilot. Integration time varies by setup, so ask the vendor what they need from your CureMD configuration.
Can we try an agent on a subset of refills first?
Yes, and you should. Start with one medication class, such as stable maintenance medications for recently seen patients, compare the agent's decisions to staff decisions, then expand as accuracy and trust build.
