Cerner automation lives where no API reaches

The two layers of Cerner automation

Almost every “Cerner workflow automation” article online describes benefits (fewer manual tasks, fewer errors) or sends you to an iPaaS connector. Both quietly assume the data you want is reachable through an API. For interoperability, it often is. For the specific chair-side and back-office tasks people are actually trying to automate, it usually is not. Knowing which layer your task sits in is the whole decision.

Why the bot you tried keeps breaking

PowerChart is a Windows thick client, usually published through Citrix. Classic RPA automates it by matching screenshots and clicking fixed coordinates. That is exactly what does not survive an EHR: a Millennium theme update, a font change, a different screen resolution, or a Citrix scaling difference between two clinics moves every pixel, and the bot misses. The team ends up re-recording the same workflow after each upgrade.

The fix is to stop matching pictures. Windows already exposes every control through UI Automation, the same accessibility interface a screen reader uses. Each field has a role (edit, button, tab) and a name (its label). Mediar locates fields by those instead of by position. You can read the exact selector syntax in the open-source Terminator SDK: github.com/mediar-ai/terminator. A step is written as role:edit|name:MRN and resolves against the live accessibility tree, the same way desktop.locator('role:edit') does in the SDK's own workflow files.

# image / coordinate RPA
- click image: "powerchart_orders_tab.png"
  confidence: 0.85
- click x: 412 y: 268        # "Add" button
- wait: 1500ms
- type: "{patient_mrn}"
- click x: 540 y: 392        # next field
- type: "{auth_number}"
# breaks when Millennium ships a new
# theme, font, or layout: every
# coordinate and screenshot is stale

The left version names a screenshot and two pixel coordinates. The right version names labels that a clinician would recognize. When Oracle reskins the chart, the label survives and the right version keeps running.

What “watch once” actually does

You do not script the workflow by hand. A user runs the PowerChart task one time. The agent records the accessibility path of every control it touches, turns that into a repeatable sequence, and from then on executes it against the live tree. There are no selectors to hand-author and no screenshots to maintain.

The cross-app task no single API covers

The most common reason a Cerner task has no API is that it spans two systems. A worker reads patient and authorization fields out of PowerChart and re-keys them into a payer portal that has its own login and its own forms. No FHIR call bridges those two. An agent that can see both UIs through the accessibility layer can: read the chart, type into the portal, then write the confirmation back into the note.

What this is worth in practice

The economics show up in patient intake and revenue cycle, where the same fields get re-typed thousands of times a week. One deployment automating patient intake removed enough manual re-keying to be measured in headcount, not minutes.

Pricing is usage-based: 0.75 dollars per minute of runtime, with a 10,000 dollar turn-key program fee that converts to credits. No per-seat license, and because the agent runs inside the existing logged-in session, there is no separate PHI export to secure. Mediar is SOC 2 Type II certified and HIPAA compliant, and runs on-prem or in your cloud.