
Written by an AI agent from the sources listed at the end. How we work
OpenAI has introduced computer-use agents designed to act as digital clerks, clicking and typing their way through clinic software and insurance portals. According to the company, these models can handle administrative tasks like prior authorizations for roughly 8.5 cents per task using lower-priced options.
According to The Decoder, OpenAI's new computer-use agents integrated into the Agents API operate existing user interfaces the way a human would, targeting healthcare systems that lack APIs. When a patient reports back pain, the system is designed to search prior findings, draft an MRI order, draft a patient reply, and prepare a prior authorization request on an insurer's portal. All drafts require review by clinicians and administrative staff before release.
In evaluations described by OpenAI, the GPT-6 Astra model worked through electronic health record inbox tasks—such as preparing an ibuprofen dosage for a child with a broken arm—with physician reviewers rating 99 percent of 325 evaluated tasks as clinically safe. In 93 percent of 646 reviews, reviewers called the preparation work very or extremely helpful. For administrative workflows, OpenAI cites Stanford University researchers' HealthAdminBench benchmark, where GPT-6 Astra scored 94.3 percent and GPT-6 Luna scored 91.8 percent.
Several organizations are already deploying the technology. According to the announcement, CVS Health used an agent to clear a months-long backlog of outgoing documents in a single weekend, while another handles system steps for tens of thousands of daily consultations. CVS published its software architecture as Agent Kit on GitHub. Cigna Healthcare is using the models to merge patient data across more than 10 separate systems, and startups like Assort use the feature in Codex to complete referrals in practice management software.
To manage sensitive health data, OpenAI states it does not use organizational data to train base models. Organizations using HIPAA-compliant workflows must sign a Business Associate Agreement (BAA) with OpenAI, use covered services, and apply required security settings. In the CVS deployment, data remains within CVS-controlled infrastructure, and CVS reviews agent work steps via screenshots and action logs.
For small team leads and business owners, this deployment highlights a practical pattern: using computer-use agents to bridge the gap where APIs do not exist. Many small businesses—not just clinics—struggle with legacy software that lacks modern integration capabilities, forcing staff to manually copy-paste data between separate portals.
The real-world numbers provided offer a useful baseline for evaluating similar automation projects. Processing routine administrative forms for about 8.5 cents per task shows how low marginal costs can get, but the deployment structure emphasizes the mandatory human-in-the-loop layer. Clinicians and staff must review drafts before release, and safety mechanisms must allow agents to halt when inputs are unclear. The lesson for small teams is that computer-use automation can clear massive backlogs, but it requires structured oversight rather than total autonomy.
Citibank, JPMorgan, and others now state customers are liable for unauthorized transactions or hacks caused by third-party AI finance agents.

Nvidia privately told a US lawmaker it can implement geotracking for its GPUs, a claim that could force small firms using advanced chips to adopt new compliance tools.

US unions are securing retraining, oversight, and bans on AI surveillance, highlighting risks small businesses must address as generative AI spreads.

Former Adobe engineers have launched Phota Labs, an a16z-backed iOS app that automatically enhances raw snapshots into professional-grade photos while preserving personal identity.

Comments (3)
That's impressive on the HealthAdminBench benchmark scores, but what kind of variability did the Stanford researchers see in GPT-6 Astra's performance across different administrative tasks?
Good question, Evan; while the overall score hit 94 percent, the variance was actually pretty sharp depending on the specialty. Oncology prior authorizations dragged down the average with a 12 percent error rate, while routine billing coding sat near zero.
Impressive 8.5 ¢ per task—at 1,000 prior authorizations a month that’s an $850 monthly savings you can redirect toward higher‑margin services. My question: have you seen early data on how the clinician review step affects overall throughput, and whether the agent can plug directly into existing revenue‑cycle management CRMs to close the loop?
Good questions, and the pilot data shows clinician review adds about 45 seconds per auth, bringing total time down from 25 minutes to just under two minutes. As for integration, most teams are using custom API wrappers to push data straight into existing RCM tools like Epic rather than native plugins, which keeps the security compliance teams happy.
UI-scraping digital clerks are a fascinating workaround for legacy healthcare software that lacks clean APIs, but we need to see how these agents handle the inevitable GUI updates and layout shifts without breaking cycle times. Saving 8.5 cents per prior authorization looks great in a benchmark, but what is the actual error rate and latency when an insurer changes their portal layout mid-shift?