Pretty Good AI vs Sully.ai

Sixteen EHRs, or one

Pretty Good AI is built for athenaOne and nothing else. Sully.ai lists sixteen EHRs on its integrations page, with athenahealth as one entry among Epic, Cerner, MEDITECH, Veradigm, NextGen and the rest, and activates named agents from a single integration layer that has to serve all of them.

That is a reasonable way to build a company. It is also the whole comparison. Every hour of integration engineering a vendor spends is spent either on breadth or on depth, and a practice feels the choice at the moment a call needs to finish.

What depth actually buys you

Voice runs under a constraint that documentation does not. There is no draft. The patient is on the line, the agent gets one pass, and whatever it decides has to be correct and has to land in athenaOne before the call ends.

That is why we hold production access to 820+ athenaOne APIs, read and write, with significant depth across the whole of athenaOne where most vendors pick a narrower wedge and build inside it. It is also the honest answer to the question practices actually ask, which is why some vendors’ calls finish and others’ calls turn into a task for the front desk. It is not voice quality. Everyone’s voice is good now. Calls finish when the agent can read the live record during the call, the real scheduling template and slot type, the appointment types a given department uses, the provider group, the insurance on file, the order and tickler queues, and write back before it hangs up.

The rule that is not written anywhere

Here is the shape of the problem a shared integration layer has to solve sixteen times.

At one vascular group, roughly nine in ten visits are a diagnostic ultrasound paired with an office visit with the physician, booked together with specific spacing between them. Patients do not know this. They call and ask for an appointment with their doctor. Nothing in the scheduling template requires the pair. The rule exists only as practice convention, which means the only way to find it is to read ninety days of their own booking history, infer the pattern, and then confirm it with the staff before writing it into a workflow.

We know exactly how that fails, because it failed on a live test call at that group. The agent booked the ultrasound and the provider visit for one body region, and the follow-up task that should have been created was not. Half the workflow silently did not exist. Staff caught it in review, which is the point: the exception surfaced to a human who was expecting to look, rather than to a patient at check-in three weeks later.

No amount of conversational quality substitutes for reading the configuration, and no generic integration layer infers an unwritten convention at one practice in one specialty. That work is specific, it is per practice, and it is what we do instead of spreading across EHRs.

Read the listing and the site as two different documents

If you are evaluating Sully.ai from inside athenahealth, one thing is worth checking before you shortlist.

Its athenahealth Marketplace listing describes the clinical assistant: notes, differential diagnosis, clinical plans, prescriptions and patient records generated from the recorded visit. The listing was last updated in June 2025 and names athenaClinicals as a required athenahealth product. Its own site today sells five agents, front desk included. Both are published by the vendor and they describe different products.

That is not a criticism, it is a question to ask on the call. Which agents are generally available for an athenaOne practice this quarter, which athenaOne records does each one write, and what does the Marketplace listing cover if you buy through athenahealth as an Alliance Partner solution rather than direct.

What runs on our side today

Voice answers and completes the work. Unanswered calls become cases in athenaOne. Secure two-way text runs on the same integration, so a thread and a call share context and write the same records. Around that: referral intake from fax, portal and web, tickler and order outreach, waitlist backfill and no-show recovery, eligibility and prior-auth prep, denials and appeals, balance outreach under your billing team’s rules, post-visit outcome calls and surveys, forms signed inside athenaOne, web scheduling and self-check-in, and per-call review with reporting on all of it. We run the full administrative call in 10 languages, English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian and Dutch, with the same workflows in each.

Clinical decisions stay with your care team. The Pretty Good AI platform has attained HITRUST i1 certification. Pretty Good AI signs a HIPAA Business Associate Agreement (BAA) before any patient data is touched. SOC 2 Type II and ISO 27001 audit reports are available upon request.

What we would tell you to check

Ask each vendor for a live call on your own tenant, not a demo tenant, and watch whether it ends with the record updated and no task created. Ask what happens on the calls that fail, who sees the exception, and who rewrites the workflow when your scheduling rules change in January.

Then ask the breadth question directly. Ask how many engineers work on the athenaOne integration specifically, and what shipped for athenaOne in the last quarter rather than for the platform generally. Our answer is all of them, and the list is long, because athenaOne is the only integration we have.

Then run the test that actually separates us. Describe the most awkward local rule you have, the one your staff work around every week, and ask each vendor how long it takes to automate that rule exactly as your practice runs it. Our answer is days to weeks inside the same onboarding window, because the depth of the athenaOne integration is what makes an unusual workflow buildable at all, and because we map the workflow your staff already run rather than asking them to re-cut it to suit the product. You do not adapt to us.

Then get the total. Ours is a platform fee per location, a per-provider fee for the modules you switch on, and voice metered per call. We agree on the first workflow, how success will be measured, and the pricing for continued use before launch. There are no setup fees. Your first 30 days live are free. The clock starts when your first agreed workflow goes live in your practice, whether that is voice, web scheduling, referrals or another workflow. When the free 30 days end, service continues automatically, month to month, at the pricing in your order form. New workflows are priced when activated. Every competitor figure on this page was checked against the vendor’s own pages and the athenahealth Marketplace on 2 October 2026.