Why coding splits in two
Medical records specialists turn clinical notes into codes, claims and clean data. Part of that job is pattern work: read a complete note, assign the right ICD-10 and CPT codes, move on. Software is good at that shape of task. Compiling record data for registries and internal reports follows the same pattern.
The rest is messier. Charts contradict themselves. A note says one thing and the lab result says another. Deciding the principal diagnosis when two conditions compete is a judgment call, and it changes what a hospital is paid. When documentation is thin, someone has to write a physician query, wait for the answer, and record it properly.
Accountability is the other half of the story. Releasing patient information under HIPAA, answering a payer audit, and defending a denied claim all need a named person who can be held responsible. That is why the honest question is not whether medical coding will be replaced by AI, but which parts of the day move to software first.
What software handles, assists with, and leaves alone
Start with the share of task time our scoring puts in the automate group: 44%. That is where straightforward code assignment from clear documentation sits, along with routine data entry and the compiling of record information into standard reports. These tasks have a defined input and a defined output, and the rules are written down.
The assisted group covers 40% of task time. Reviewing records for completeness and accuracy fits here: the tool flags gaps, the specialist decides whether the gap matters. Tracking charts through the review cycle works the same way, with software queuing the work and a person judging the exceptions.
Work our scoring leaves with a person comes to 16% of task time. That is the physician query, the appeal of a denied claim, and the release-of-information decision where privacy rules and a real request have to be weighed together. Overall coverage reads 54 out of 100; the coverage method page explains what that counts.
What has actually been tested
Not much, directly. The evidence grade for this job is D. A D grade means there is no published head-to-head test of AI against qualified medical records specialists on the same charts, so we give no parity number at all. Vendor claims about autonomous coding accuracy are not the same thing as an independent comparison.
What would settle it is simple to describe and rare to see: a blind audit on a shared set of real charts, inpatient and outpatient, comparing coder output with software output, with coding accuracy, query rates and downstream denial rates all reported. Until that exists, the sensible reading is that software performs best on clean, short, single-issue documentation and worst where the chart is incomplete. The quality parity method sets out how we grade that kind of test.
Good to know: the Bureau of Labor Statistics counts about 194,720 people in this occupation with median pay of $51,140, and projects 7.8% growth between 2025 and 2035 (BLS, 2025).
What could move the date
Most likely between 2039 and 2049 (8 in 10 of our scenarios). The replacement year method explains how that window is built.
Two things could pull it earlier. First, no hardware is involved. This is screen work, so there is no robot to build, install or maintain, which removes the slowest step most occupations face. Second, the cost gap shown in the panel above is wide, and large hospital systems and billing companies process enough volume to justify buying once and running everywhere.
Two things hold it back. Payer audits and documentation rules mean errors are expensive and traceable, so health systems keep human review in the loop even when software drafts the code. And the untested quality question matters here: without an independent comparison, compliance officers have little ground to sign off on unreviewed output. Entry-level roles feel this first, because the simplest charts are the ones software clears without help.
How to stay needed in this job
Lean into the work that sits in the human group above. Own the physician query process, including how queries are worded so they are compliant and actually get answered. Take the denial and appeal work, where you have to read a chart, a payer policy and a rejection together. And handle release of information, where privacy law and judgment meet.
Two skills raise your floor. One is audit and quality review: checking AI-suggested codes against documentation and explaining, in writing, why a code was changed. The other is specialty depth, especially inpatient and risk-adjustment coding, where documentation is long and the rules shift.
Nearby roles worth a look if you want to move sideways: Health Information Technologists and Medical Registrars, Medical Transcriptionists and Billing and Posting Clerks. You can also see how this job sits against others in the health technologists and technicians family or across the healthcare sector, put two jobs side by side on the compare page, or check the jobs most at risk list. Every score on this page comes from open data and a published scoring method.