what changed for clinicians of any background building their own software with ai — coding agents, models, health data access, privacy & the rules — checked daily, each note against at least two sources
What a clinician needs in order to build software, in 8 layers, and where each one stands. Every note belongs to one or two.
A coding agent is a program that reads your project, writes and runs code, and fixes its own mistakes while you describe what you want in plain language. This is the layer that lets a clinician with no engineering team build working software. In late 2026 the agents themselves became a target: a tampered Claude Code extension in Cursor's marketplace, a hijacked Microsoft Python package, and flaws in the built-in browsers of five AI coding tools. Choosing an agent now also means watching what it installs and what it can read.
Latest: OpenAI releases GPT-6.2 Astra at $8/$40 and GPT-6.2 Sol at $2/$10
The hosted language models behind the agents, and the APIs your own app calls when it needs to read, summarize or reason. Prices for top models fell through autumn 2026, with Anthropic and OpenAI both cutting rates. What matters to a builder is capability on your own task, price per use, and the terms the vendor will sign. A privacy setting may bind only the provider you send to directly, not resellers further down the line.
Latest: Canary study finds AI providers and resellers exposed prompts and keys
Open-weight models you can download and run on your own hardware, so patient data never leaves the building. They trail the hosted models in raw ability, but published clinical studies testing them on real tasks, such as flagging urgent patient messages, are starting to appear. Running locally removes one privacy risk, not your other duties.
Latest: Study compares cloud and in-house open-weight models for flagging urgent portal messages
How your software gets at clinical data: FHIR, the standard format most EHRs share records in; SMART on FHIR, the way an app signs in to an EHR; vendor developer programs; and the rules on who is entitled to what. Epic plans to let outside tools plug into its own AI agents through Connectors built on MCP, an open standard for connecting agents to tools, from March 2027. A HIPAA Privacy Rule update finalized in November 2026 lets patients have their records sent to an app they control, where the covered entity can provide them that way. Federal certification rules for EHRs were narrowed in October 2026, but the information-blocking rules stay in force.
Latest: HHS finalizes HIPAA Privacy Rule update: 15-day access deadline, health app copies
What has to be true before real patient data touches anything you build: which vendors will sign a business associate agreement (BAA), what they retain, what HIPAA requires of software, and the security of the tools themselves. Late 2026 brought a run of attacks aimed at AI developers' keys and machines, and a study showing prompts leaking through AI resellers. Keep secrets out of project files and prompts, and keep real PHI off machines where you run agents on code you did not write.
Latest: Canary study finds AI providers and resellers exposed prompts and keys
The line between a helpful tool and a regulated medical device, and the federal and state rules on AI in care. FDA's clinical decision support guidance draws the federal line. States are adding their own: California's AB 1979 and SB 503, effective January 1, 2027, set duties for clinical AI and its developers and deployers. Federally, ONC dropped its AI transparency requirements for certified EHRs in October 2026.
Latest: ONC finalizes HTI-5, dropping 14 certification criteria and AI transparency rules
Where the thing you built runs once it works on your laptop: hosting that will sign a BAA, app platforms and no-code builders, and the ways to share a tool with colleagues or put it inside an EHR. No notes yet; this layer fills as developments land.
What physicians, nurses, pharmacists and informaticists have actually built and deployed, the published evidence on how it went, and the courses and programs teaching clinicians to do it. Short accredited AI courses aimed at clinicians are beginning to appear.
Latest: Study compares cloud and in-house open-weight models for flagging urgent portal messages