AKASA launches autonomous AI for inpatient coding and documentation

AKASA said Friday it launched an autonomous AI platform for inpatient medical coding and clinical documentation integrity, expanding from AI prebill review into mid-cycle work that used to take human coders 30 to 60 minutes.

Hospitals still turn clinical charts into billable codes largely by hand — and error rates plus coder shortages make that a bottleneck. AKASA’s bet is that generative AI fine-tuned per health system can finish the hardest inpatient coding without a human in the loop, cutting days of A/R while keeping a compliance-first path that systems like Cleveland Clinic are willing to explore.

On Friday, 2 October 2026, AKASA said it launched an autonomous AI platform for the healthcare mid-cycle, covering inpatient medical coding and clinical documentation integrity. The PR Newswire page is titled “AKASA Launches First Autonomous AI Platform for Coding and Documentation.” The line under the headline says the company’s customer base represents 1 in 10 U.S. inpatient discharges as it launches the autonomous mid-cycle for inpatient coding and clinical documentation, in what the page calls healthcare’s most complex and resource-intensive workflows. The page stamps Oct 02, 2026, 07:05 ET. The dateline is South San Francisco, California. The source line is AKASA. A revenue cycle is the path from a hospital stay to the bill. The mid-cycle is the middle of that path, where the chart becomes the codes. Clinical documentation integrity, often shortened to CDI, is the work of making the chart complete and specific enough to support those codes. Prebill review is a check of the claim before it goes out. AKASA says it is expanding from AI-powered prebill review into coding and CDI that can run without a person finishing the chart. “First” is the headline’s word. Those lines are the wire’s.

What the release says that middle step is for. The mid-cycle is where a patient’s clinical record is translated from the notes into codes that drive reimbursement, quality reporting, risk adjustment, and the integrity of the patient record. Reimbursement is what the hospital is paid. Quality reporting is the set of scores a hospital sends about how care went. Risk adjustment is the estimate of how sick the patients are, which changes payment. The release says the work is highly complex, uses a lot of staff time, and is still mostly manual. It cites a 2025 peer-reviewed study in the journal npj Health Systems for medical coding error rates of up to 20 percent, about 1 chart in 5. It also cites a July 2026 review by the U.S. Government Accountability Office that names verifiable accuracy as a central challenge when organizations use AI for medical notes and coding. Those two citations are the release’s. They are not a score AKASA printed for Friday’s platform.

What the company says the new platform does. AKASA says the technology is designed to fully code highly complex inpatient cases across all specialties with no human intervention. Inpatient means the patient was admitted, not treated and sent home the same day. A specialty is a branch of medicine, such as cardiology or orthopedics. The company says it will soon add outpatient facility encounters, the visits and procedures that do not include an overnight stay. The release says the platform is meant to expand the workforce’s capacity, speed billing, and improve quality performance. “Designed to,” in those sentences, is the company’s description of the aim. Beyond coding, the release says AKASA extends the same tools upstream into CDI, so documentation and coding sit in one AI layer with the prebill review the company already sells. As it rolls the mid-cycle out, the release says, it will work with health systems on deployment plans that scale how much volume runs on its own. Those lines are AKASA’s, on the wire.

When a hospital can turn the coder on, as Fierce Healthcare reports it. Heather Landi’s story, dated Oct. 2, 2026, at 7:30 a.m., says AKASA gave the outlet a first look at the platform. Malinka Walaliyadde, chief executive and co-founder, told Fierce that after internal testing and an evaluation with a health-system alpha partner, AKASA plans to make the autonomous platform for inpatient medical coding available in the next several months. An alpha partner is an early customer trying the product before a wider release. That timeline is Fierce’s account of the interview. The wire calls Friday’s news a launch. It does not print “next several months.”

The speed claim, kept as the company’s. The wire says a coder typically takes 30 to 60 minutes to code one inpatient encounter. Shortages can leave a health system waiting several days after discharge before a coder even starts the account. AKASA says its AI can finish the coding in less than 90 seconds after the patient is discharged. Ninety seconds is a minute and a half. Thirty to 60 minutes is half an hour to an hour. The release says the shorter wait speeds billing and cuts accounts-receivable days. Accounts receivable, often shortened to A/R, is money billed and not yet collected. An A/R day is one day that money sits uncollected. Fierce’s story says “approximately 90 seconds,” the same company claim in the reporter’s wording. The pages do not print a stopwatch from a named hospital.

How big one chart is, as the chief executive described it to Fierce. Walaliyadde said a typical inpatient stay is about 60 documents and 50,000 words, and that a coder turns that stay into codes chosen from a set of about 150,000. Those counts are his, in the interview. On the wire he is quoted: “The incredible demand for healthcare is finally being addressed by advancements in AI. Multiple parts of the healthcare ecosystem will need to scale up, with documentation and coding being critical components.” He added: “For years, an autonomous mid-cycle has been a holy grail in our industry. Today, AKASA is making it real.” Fierce also quotes him on a coming squeeze: more care, harder care, and not enough people to code it, and on his view that AI can now do that work. “Holy grail” and “making it real” are his words. The wire does not print the 60-document count. Fierce does.

The customer-base figures, kept as AKASA’s. The wire says inpatient volume the company processed grew almost 6 times in the last year. It says customers represent more than $180 billion in aggregate net patient revenue and roughly 10 percent of U.S. inpatient discharges. Net patient revenue is the money a hospital expects to collect for care, after the discounts insurers negotiate. Aggregate means the customers’ totals added together. Ten percent is about 1 in 10 discharges, the figure in the subhead and on the company homepage. Six times is the company’s line for volume it processed. It is not a claim that the software codes every chart at those hospitals, and it is not a claim about every discharge in the country. The homepage also says the client base represents 500 hospitals across all 50 states. That hospital count is the site’s. The wire does not print it.

How the models are built, as the company states it. AKASA says it fine-tunes a model for each health system, so the software can account for that system’s patients, its clinical criteria, the way clinicians write, and how complex the care is. The about box says the models are trained on that system’s own clinical documentation, its case mix, and its coding decisions. A case mix is the blend of illnesses a hospital treats. The homepage says each health system gets its own model, and that the system sets the threshold: how much volume runs without a person follows that system’s case mix and quality standards. Walaliyadde told Fierce that a generic model does not hold up on this work, and that the custom model then powers several applications at that health system. Those lines are AKASA’s. The pages do not name the base model.

The accuracy claim, kept as AKASA’s account of its blinded evaluations. The wire says the company ran third-party blinded evaluations that compared the AI with medical coders. Blinded means the people scoring the work were not told which codes came from the software. Third-party means AKASA says the review was not only an internal demo. In those studies, the company says, it tested inpatient encounters representing 65 to 85 percent of inpatient volume for health systems, and the AI matched or exceeded human coders on four measures: MS-DRG assignment, principal diagnosis, clinical quality capture, and present-on-admission accuracy. An MS-DRG, a Medicare Severity Diagnosis Related Group, is the payment bucket for a hospital stay. The principal diagnosis is the main reason for the stay. Present on admission means the condition was already there when the patient arrived, which changes how a complication is paid and counted. Clinical quality capture is whether the codes pick up the quality measures the record supports. Sixty-five to 85 percent is about two-thirds to a bit more than four-fifths of volume in that test, as the company describes the sample. The homepage prints a different sentence beside a “matched or exceeded expert coder accuracy” line: “65 to 85% of inpatient volume at most health systems autonomously addressable,” and it says the evaluation ran “across entire code sets.” Addressable, on the site, is the company’s word for volume it says the software can take. The wire’s 65 to 85 percent is the share of volume in the encounters it says it tested. Fierce repeats the company’s evaluation account. It does not publish the raw scores. The match-or-exceed line is AKASA’s account of those evaluations.

Cleveland Clinic, as the pages state it. The wire says the clinic already uses AKASA’s prebill products across coding and CDI, and that it intends to explore autonomous mid-cycle solutions with AKASA, to strengthen the accuracy and completeness of the record and to improve operations. Rohit Chandra, Ph.D., chief digital officer, is quoted: “Our revenue cycle work is especially time-intensive because we care for many medically complex patients. With autonomous coding, we seek to improve speed and precision in these challenging processes under a compliance-first approach to this work.” That quotation is his. Fierce’s article says the clinic plans to deploy the autonomous tools, and it carries the same Chandra quotation, given to Fierce as a statement. The wire’s verb is explore. Chandra’s sentence is “we seek.” Neither page prints a start date, and neither page says the autonomous coder is already finishing Cleveland Clinic’s charts.

Two more quotations. Jeff Francis, chief financial officer and vice president of finance at Nebraska Methodist Health System, said on the wire that AKASA has been a pioneer in generative AI and in revenue-cycle tools, that his system has moved with the company step by step, and that the results he watches are revenue integrity, denials, write-offs, and how quickly the system gets paid. He called autonomy the natural next step as systems look for capacity. Generative AI, here, is a model that reads a record and produces the coding work, rather than only a fixed list of rules. Fierce says Nebraska Methodist has worked with AKASA since 2019. The year is Fierce’s. The wire says several years and does not print 2019. Julie Yoo, a general partner at Andreessen Horowitz, the firm often shortened to a16z, said on the wire that healthcare cannot meet the demand ahead through human labor alone, and that AKASA’s mid-cycle technology shows what is possible when AI is built for complex documentation and inpatient coding. Fierce describes her as a primary backer of the company. In that interview she called the inpatient product “truly first in class,” and said most startups in this market start on outpatient cases, which she called a simpler mix and a lower-stakes use. “First in class” and that comparison are her words, in Fierce. The wire does not say a16z invested. Fierce does.

Who the about box says AKASA is. The company says it builds AI for the hardest work in the healthcare revenue cycle, and that a health system can start with tools that help CDI and coding teams and go as far as workflows that code an encounter from start to finish. The homepage’s banner reads “The autonomous mid-cycle is here,” and a label says the news was announced at AHIMA26. AHIMA is the American Health Information Management Association, the group for people who manage health records and coding. The site also lists products already on the line: a Prebill Optimization Suite, a Coding Optimizer, a CDI Optimizer, an AI Advisor, and tools that check whether a procedure was authorized and where a claim stands. Those names are the site’s map of the products the launch sits beside. The site prints a phone number, 650-209-0358, and info@akasa.com. The release does not print a price.

The picture is a coral launch graphic. A navy panel on the left reads AKASA, then AUTONOMOUS MID-CYCLE, then INPATIENT CODING + CDI. On the coral field, large type says <90 SEC, with the line “AI coding after discharge vs 30–60 min for a human coder.” A second block says 1 IN 10 and “U.S. inpatient discharges in AKASA’s customer base.” The footer reads AKASA / PR NEWSWIRE · OCT 2, 2026. The date on the graphic is that credit line. The frame is not a photograph of a hospital. It does not show a chart, a price, or a regulator’s mark.

In plain terms, AKASA said on Friday, from South San Francisco, that it is moving from AI that reviews a bill before it goes out to software meant to code complex inpatient stays, and to support the documentation behind those codes, without a person finishing the chart. The company says that coding can finish in under 90 seconds after discharge, against 30 to 60 minutes for a human coder, and that shortages can leave accounts waiting days. It says customers account for more than $180 billion in net patient revenue and about 1 in 10 U.S. inpatient discharges. The accuracy comparison is AKASA’s account of third-party blinded evaluations on encounters it says represent 65 to 85 percent of inpatient volume. Cleveland Clinic, already on the prebill products, says it intends to explore the autonomous tools. Fierce Healthcare reports that the chief executive put wider availability in the next several months, after an early health-system partner. The pages do not print a price, and they do not describe a regulator’s clearance.