Congratulations to @OutcomesAI, winner of the Clinical Decision Support Systems category in Newsweek's inaugural AI Impact Health Awards.

But Continuum goes beyond decision support. Every suggestion its AI makes arrives as a draft that the surgeon who did the work reads before it becomes part of the record.

The last suture is in, and the operative note is already writing itself on the screen. The approach. The implants logged during the case. The medications from the anesthesia record.

Then the surgeon stops on one sentence. It describes the closure in words the surgeon would never have used.

The draft waits. It can't enter the chart until the surgeon finishes reading.

When software suggests something that isn't quite right, who catches it?

Regulators have thought hard about that question. On September 28, 2022, the FDA published its final guidance on Clinical Decision Support Software. Its fourth criterion describes software built so a clinician can independently review the basis for its recommendations and relies on their own judgment rather than primarily on the software.

That principle cuts differently depending on how a product is built. A model that reads an entire chart and proposes a diagnosis asks the clinician to check a conclusion it reached by probability. A draft assembled from structured data the clinician can inspect line by line is far easier to verify. That's an inference about design rather than a regulatory ruling, but it matters for patients, because errors that are easy to see tend to get caught.

Continuum is built on the second approach. It keeps the perioperative record from pre-op through anesthesia, the procedure, implants, medications, recovery and discharge, and it streams operative-note drafts from that record for the surgeon to review. The rest of the network follows the same rule. Revela drafts operative notes alongside a surgical-safety screen. Mind raises rating-scale safety alerts for the clinician to evaluate. Forge suggests schedules that staff can accept or ignore.

Underneath sits a structured clinical knowledge layer that traces back to a clinician-designed terminology begun in 1998. The AI drafts against defined clinical concepts, not just free text, which gives every draft something firm to be checked against.

There's a quieter point here for anyone thinking about where lasting value sits in clinical AI. Software that decides carries the weight of every decision it makes. Software that drafts for review leaves a clear trail instead: what it proposed, what the surgeon changed, who approved the final version and when. That trail is what hospital counsel, malpractice carriers and compliance officers look for before they let AI spread across an organization. It's also what lets the same assistance reach every specialty without moving judgment away from the people licensed to exercise it.

Generative diagnosis is a genuine frontier, and some of it will earn its place in medicine.

But if you're trying to judge which AI will make care safer rather than just faster, watch the surgeon's pause. Who is the last reader, and does the record show it?