Fast AI Workflow Layers Over EHRs
Caroline Zhang, CEO of Knowtex, on going from AI medical scribe to AI lab
This opened the door for AI startups to become the real pace setter in clinical software, even while EHRs still own the record system. Hospitals can keep Epic or Oracle as the database of record, then buy a faster moving layer on top that handles chart prep, ambient capture, coding, orders, and trial matching. That is why companies like Knowtex can win without replacing the core EHR, they just have to plug into it and ship useful workflow upgrades much faster.1 2 3 4
- The gap is product velocity. Knowtex describes shipping new workflow features in months, while health systems are often still stuck in long EHR upgrade cycles and migrations. In practice, that means a startup can add useful outputs like billing codes or order drafts before the incumbent rolls out its next native module.1 4
- The market already rewarded this faster overlay model. Abridge scaled from about 8,000 to more than 60,000 clinicians across 100 plus health systems after pairing strong workflow software with EHR distribution, and Ambience also expanded from note generation into broader revenue and clinical workflows. The winner is no longer just the best note taker.2 3 5
- This also explains why standalone scribing weakened so quickly. Once free or bundled note tools showed up in OpenEvidence, Doximity, Epic, and others, startups had to move into the harder parts of the workflow that EHRs and freebies still handle poorly, like previsit chart review, specialty specific coding, and follow up actions inside the care process.1 5
Going forward, the most valuable healthcare AI companies will look less like note apps and more like fast moving workflow layers on top of legacy records. EHRs will remain the system of record, but startups that turn patient context and live conversations into actions, decisions, and new training data will keep taking the innovation lead.1 4 5