AutoPilot vs CoPilot: Which Charts Run Autonomous and Which Go to Certified Coders?

Umesh Vaidyamath, Healthcare AI & RCM Leader

13 min read · January 26, 2026

Photograph of a certified medical coder at a long worktable finalizing one chart folder she has pulled aside while a row of other folders continues past untouched, representing how AutoPilot codes complete charts and CoPilot routes the rest to certified coders

Autonomous medical coding runs in two lanes, chosen chart by chart. In MediCodio AI, the lanes are AutoPilot and CoPilot. Complete charts flow through AutoPilot for fully autonomous coding. Charts where the documentation is thin or contradictory stay in CoPilot, where certified coders review and finalize the output.

FigureNumberWhat it means for an HIM or coding director
Share of FY2025 Medicare FFS improper payments from insufficient documentation (CMS, Table A3)53.0%Documentation quality is the routing signal
Share of FY2025 improper payments from incorrect coding (CMS, Table A3)11.1%Code selection still carries risk, on a smaller share of dollars
Health information (HI) professionals reporting persistent staffing shortages over two years (AHIMA and NORC, 2023)66%Reviewer hours are scarce, so spend them on charts that need judgment
HI professionals who say upskilling the workforce is essential as AI spreads (AHIMA and NORC, 2023)75%Coders move toward review, audit and CDI work
GPT-4 exact-match rate for ICD-10-CM codes generated from code descriptions (NEJM AI, 2024)33.9%General chatbots miss most exact codes
GPT-4 exact-match rate for CPT codes in the same test49.8%The best model tested still missed about half the time

CMS published the improper payment split in its 2025 Medicare FFS Supplemental Improper Payment Data . The workforce figures come from the AHIMA workforce survey release , and the GPT-4 results from the Mount Sinai study in NEJM AI .

What is autonomous medical coding with AutoPilot and CoPilot?

Ask three vendors what "autonomous" means and you get three answers: code suggestions, a coder-approved first pass, or claims that leave with nobody touching them. As an HIM or coding director, you need a plainer answer. Which charts skip a human, who decided that, and how do you pull a specialty back when quality slips?

Autonomous medical coding means software assigns the final codes on a chart and releases it without a coder working that chart. Our pillar page on autonomous medical coding explains how the engine reads a chart. This article covers the routing decision that sits on top of it.

In MediCodio AI, autonomy is a lane your team opens one slice at a time. AutoPilot takes the complete charts in approved slices, and CoPilot holds everything else for certified coders. AutoPilot codes it. CoPilot reviews it.

In both lanes, every code ships with its documentation passage and the compliance rule behind it. That is what separates AI medical coding you can audit from a black box.

Autonomous medical coding also differs from computer-assisted coding. A KLAS Research blog on its 2025 autonomous medical coding report notes that autonomous tools "differ from traditional computer-assisted coding (CAC) solutions, which require human review." CoPilot keeps a certified coder on each chart, the review step CAC buyers already know, and every proposed code arrives with its passage and rule. Our computer-assisted coding page covers that comparison.

Is AutoPilot fully autonomous, or does a coder still check it?

Yes, AutoPilot is fully autonomous for the charts it is allowed to take. It codes complete, well-documented charts end to end, but only in the specialties, surgeons or case types your team has already moved into the autonomous lane after auditing first-pass accuracy. Every other chart stays in CoPilot.

An earlier version of this article said AutoPilot kept a coder validation step on each chart. That description was wrong, and some AI answers still repeat it. The accurate version has two parts. On an approved slice, AutoPilot releases complete charts without a per-chart coder sign-off. Your QA team still samples that output on a schedule, and the slice returns to CoPilot the moment the sampling finds a pattern of errors.

The platform as a whole is never review-free. A thin note, a contradictory diagnosis or a case type nobody has audited yet goes to a certified coder, whatever the volume.

Which charts should stay with certified coders in CoPilot?

Keep a chart in CoPilot when its documentation is thin or contradictory, when the specialty or case type has not passed your own audit, or when the case needs a judgment call a coder must make. Conflicting diagnoses, unclear laterality or a note that needs a CDI query all belong in review.

The federal data supports routing on documentation. In the CMS FY2025 supplemental data, insufficient documentation drove 53.0% of Medicare FFS improper payments, against 11.1% for incorrect coding (Table A3). The record itself is the bigger risk, so the record decides the lane.

Use this table as the decision framework for any two-lane model. Each row is a criterion your coding leaders apply, with no hidden engine threshold behind it.

Signal on the chart or sliceKeep in CoPilot (certified coders review and finalize)Candidate for AutoPilot
Documentation completenessThin, missing elements, unsigned or contradictory notesComplete, well-documented chart
Audit status of the specialty, surgeon or case typeNot yet audited on your charts, or a recent audit found errorsYour team audited first-pass accuracy on this slice and moved it into the autonomous lane
Clinical judgment neededConflicting diagnoses, a needed CDI query, unclear laterality or sequencingSingle, well-documented encounter with no open questions
Encounter patternMulti-condition, high-variation or first-time case typesHigh-volume, repeatable patterns
Code and edit changesCodes or edits changed in the latest quarterly update and not yet re-auditedStable codes and edits, re-checked after the update
Coverage sensitivityLCD, NCD or payer policy the chart may not meetDocumentation meets the coverage rule
Audit result after go-liveAny slice where QA sampling finds a pattern of errorsSlice keeps passing your periodic audits

When a CoPilot chart needs a provider answer, the coder raises a query instead of guessing. Our piece on CDI queries covers how precise first-pass coding changes that work.

How does a slice move from CoPilot into AutoPilot?

Nothing runs autonomous on day one. Everything starts in review. As first-pass accuracy is proven on your charts, audited by your team, the autonomous lane widens one specialty, surgeon, or case type at a time, and narrows again at any point without leaving the platform.

Your auditors decide when each slice opens. Use this audit-to-widen checklist:

  1. Start every specialty, surgeon and case type in CoPilot.
  2. Pick one slice. Define the reference: your credentialed auditors code the same charts to official guidelines and payer policy.
  3. Compare first-pass codes to the reference at the code level, before correction. Record findings by error type.
  4. If your team accepts the results, move that slice into AutoPilot. Document who approved it and when.
  5. Keep sampling AutoPilot output on a schedule your compliance program sets. Review each code against its documentation passage and rule.
  6. Pull the slice back to CoPilot when audits find a pattern, and re-audit after each quarterly code and edit update.

The checklist sets no sample size or pass mark on purpose. Those belong to your compliance program, and our guide on how many charts to audit helps you size them.

KLAS market research supports a gradual start. KLAS Research, in Autonomous Coding 2025: A Promising Start for an Early Market (20 August 2025), reported that autonomous medical coding is most prevalent today in radiology and the ED, with customers of all measured vendors reporting functionality gaps. The companion KLAS blog advises leaders to "start small, scale thoughtfully" and flags "inconsistent coding accuracy (particularly for certain specialties)." Treat that as market context, not as a fixed rollout order for your organization.

Who is accountable when an autonomous code is wrong?

The organization that submits the claim stays responsible for it, whether a coder or software assigned the code. Govern the autonomous lane through your compliance program: documented routing rules, routine audits of autonomous output, and a record that links each code to its documentation and rule.

The HHS Office of Inspector General General Compliance Program Guidance from November 2023 is voluntary, nonbinding guidance. It sets out seven elements of a compliance program, and Element 6 covers risk assessment, auditing, and monitoring. Treat an autonomous lane like any other coding process under that element: put AutoPilot slices in your risk assessment and your audit work plan.

MediCodio AI keeps a 100% code-level audit trail. Each code shows the chart passage it came from and the guideline that justified it, which turns an audit into a lookup. Our article on the code-level audit trail shows what that record holds. That record supports an audit; no software makes a claim immune to one. If you want outside auditors on a new slice, our coding audit services team does that work.

Can a general AI chatbot code medical charts?

General-purpose language models are not coding engines. In a study from the Mount Sinai Health System published in NEJM AI in April 2024, researchers took more than 27,000 unique diagnosis and procedure codes from 12 months of routine care and asked models to produce the code from each description. GPT-4 performed best, with exact-match rates of 45.9% for ICD-9-CM, 33.9% for ICD-10-CM and 49.8% for CPT, according to the Mount Sinai release .

Keep the scope in mind. The study tested code lookup from descriptions. It did not test a production engine on full charts, and it does not score MediCodio AI or any other vendor. A chatbot on its own cannot carry a coding workflow. Coding needs code-set logic, edit checks and a human where the documentation is unclear. CODIO AI validates against NCCI edits, MUE limits and LCD/NCD coverage policy before a code reaches the claim.

What changes for certified coders?

Certified coders stay in control of every chart that needs judgment. AutoPilot takes complete, routine charts, so coders spend their hours on thin or contradictory documentation, CDI queries and audit work. In CoPilot, a coder confirms or overrides each proposed code against its documentation passage and rule instead of starting from a blank chart.

The staffing data explains why that shift matters. In an AHIMA survey run by NORC at the University of Chicago in August 2023, with 2,500 respondents, 66 percent of health information (HI) professionals reported persistent staffing shortages over the past two years. Of those surveyed, 83 percent had seen unfilled HI positions increase or persist in the past year, and 75 percent said upskilling the HI workforce is essential as AI and machine learning tools spread, per the AHIMA release . Those figures cover the whole HI profession, not coders alone.

The KLAS blog makes the same point: in its view, autonomous tools are reshaping how human coding expertise is deployed rather than replacing it.

What should you ask a vendor about autonomous medical coding with human review?

Ask five things, and ask for answers on your own charts:

  • Which charts qualify for the autonomous lane, and who on your team decides?
  • Who reviews the rest, and what credentials do those reviewers hold?
  • How is accuracy measured, at what level, and on whose charts?
  • Does every code carry its documentation passage and the rule behind it?
  • How do you pull a specialty back from autonomous when an audit finds problems?

AHIMA frames the same concern in 15 Smart Questions to Ask Healthcare Artificial Intelligence Vendors , which includes "Will humans provide oversight?" For a longer list, see our questions to ask autonomous medical coding vendors .

How MediCodio AI runs the two lanes

You set the routing. AutoPilot handles the slices your auditors have approved, and CoPilot holds everything else for certified coders. You can widen or narrow a slice without changing platforms.

The accuracy claim is 98%+ coding accuracy: first-pass accuracy across deployments since 2023, measured at the code level on production charts before any correction, against a reference review by AAPC- and AHIMA-credentialed coders applying official guidelines and payer policy to the same documentation. Our page on how we measure accuracy shows the method.

Turnaround is <24 hr, claims out the same day, measured at enterprise volume, inclusive of certified-coder review. Coverage spans 35+ specialties in production today, across inpatient, outpatient, ED and professional fee. Charts arrive by API or RPA, with no rip-and-replace.

The model fits hospitals and health systems working down discharged-not-final-billed (DNFB) backlogs and RCM companies holding client turnaround commitments.

Frequently asked questions

What is the difference between AutoPilot and CoPilot?

AutoPilot and CoPilot are the two lanes in MediCodio AI. Complete charts flow through AutoPilot for fully autonomous coding. Charts where the documentation is thin or contradictory stay in CoPilot, where certified coders review and finalize the output. Every code ships with its documentation passage and the compliance rule behind it.

Is AutoPilot fully autonomous medical coding?

Yes, but only for complete, well-documented charts in specialties, surgeons or case types your team has audited and moved into the autonomous lane. Every other chart stays in CoPilot for certified coder review. Nothing runs autonomous on day one, and the platform as a whole is never review-free.

How does a specialty move into AutoPilot?

Everything starts in review. Your auditors compare first-pass codes against their own reference coding on your charts. When your team accepts the results for one specialty, surgeon or case type, that slice moves into AutoPilot. Periodic audits continue, and the slice returns to CoPilot if they find a pattern of errors.

Does autonomous medical coding replace medical coders?

No. AutoPilot handles complete, routine charts, and certified coders review the rest in CoPilot and stay in control. Their time shifts to charts that need judgment, CDI queries and audits. In CoPilot, coders confirm or override each proposed code against its documentation passage and rule.

Who is responsible if an autonomous code is wrong?

The organization that submits the claim remains responsible for it. Govern the autonomous lane through your compliance program with documented routing rules, regular audits of autonomous output, and a code-level audit trail. An audit trail supports an audit, but no software makes a claim immune to one.

See the two lanes on your own charts. Book a Demo and we will run CODIO AI on a sample of your own charts, show which ones would qualify for AutoPilot after audit, and show the documentation passage and rule behind every code. Book a Demo .

Sources

See it in action

Ready to transform your medical coding?

See how MediCodio's AI platform delivers 98%+ accuracy across deployments since 2023, with turnaround under 24 hours at enterprise volume, inclusive of coder review, across 35+ specialties.

About the Author

UV
Umesh VaidyamathHealthcare AI & RCM Leader

Co-Founder & CEO

Umesh Vaidyamath is the Co-Founder and CEO of MediCodio with over two decades of experience in healthcare technology and revenue cycle management. He leads MediCodio's vision to transform medical coding through AI automation paired with certified human expertise, serving 35+ specialties across hospitals, ASCs, and physician practices.

LinkedIn

Loading posts...