Camunda

Agentic Orchestration

Every process in healthcare is legacy

AI spend in healthcare keeps climbing while time to therapy and cost to serve barely move. The reason is not the agents. It is that every process they are dropped into was built for a world without AI, and only re-engineering the work turns pilots into production.

By Natosha McNeal

AI changed the workforce. Now change the workflow.

A specialty drug gets approved for a patient on Tuesday. The AI agent your team just rolled out drafts the prior authorization in seconds, pulls the clinical notes, and flags the right billing code. Impressive. Then the case sits. It waits on a benefits check in one system, a medical-necessity review in another, a signature living in someone's inbox. By the time the therapy clears, it is the following week, and the agent that saved you ninety seconds saved you nothing that counted.

That pattern is everywhere in healthcare and life sciences right now. Spend on AI keeps climbing. The numbers leadership actually watches, time to therapy, cost to serve, cycle time, barely move. The agents are not the problem. The processes they get dropped into are.

Every process you run was built for a world without AI

Here is the uncomfortable part. Every core process running your organization today is legacy. Not because it is old or badly built, but because it was designed around constraints that no longer apply: manual review because there was no other way to check the work, paper and fax because that was the medium, handoffs between systems because nothing coordinated them. Those constraints shaped the process. The constraints have expired. The process has not.

So when you add an agent to that process, you speed up one step inside a structure that was never built to move faster. The gains stay incremental, roughly twenty percent, and then they flatten. Camunda's 2026 State of Agentic Orchestration and Automation research puts hard numbers on it: 71% of organizations are using AI agents, and only 11% of use cases reached production last year. The other 89% are stranded between a demo that worked and an operation that runs. Separately, 72% of organizations say a process problem has already killed an AI initiative, at an average cost of $1.55 million each.

Why healthcare feels it first

Healthcare and life sciences hit this wall harder and sooner than most industries, for reasons that will sound familiar.

Your work is regulated from end to end. An agent cannot quietly approve a claim, adjust a protocol, or file a safety report without someone being able to show a regulator exactly why it happened. Your processes run long and cross many hands: a single claim or clinical case can touch a dozen systems and several teams over days, weeks, or months. Your workforce is thinning, with experienced reviewers and adjusters retiring faster than they are replaced. And the rules keep multiplying and diverging, from FDA guidance on AI in regulated decisions to CMS interoperability mandates to the EU AI Act and GxP.

Put those together and the instinct is to rip out the core systems and rebuild for AI. Resist it. Core replacements in this industry run for years, and the business cannot wait years. The answer is a different target, not a bigger project.

The fix is re-engineering the process, not a smarter agent

Stop automating one task at a time and hoping the steps connect. Re-engineer the whole journey, and put AI to work inside it with real authority and real controls.

That means three things working together. Agents handle the volume: the routine prior authorizations, the standard intake, the first-pass document review. People handle the judgment: the complex case, the appeal, the exception. And the process itself holds the line. Before an agent issues a determination, pays a claim, or submits a report, an enforceable step runs first: a medical-necessity check, an approval threshold, a compliance gate. The agent reasons; the process decides what it is allowed to do with that reasoning. That is what lets you give an agent genuine authority over regulated work without losing control of it. Camunda coordinates the agents, people, and systems across the full journey (outside the agent) and governs the steps between an agent's decision and its action (inside the agent), in one place.

Two things make this safe to run in your environment. Every decision, every path, every exception is recorded as the work happens, so when a regulator or an auditor asks why a determination was made, the answer already exists. And none of it requires replacing what you have. The coordination layer sits above the systems you already run, your EHR, your policy admin, Veeva, Oracle, so you keep what works and change what does not. People call this agentic orchestration. The label matters less than what it does, which is turn pilots into production.

What it looks like in production

This is not theory, and it is already the 11% that got past the demo.

Evernorth put specialty-pharmacy prior authorization on an orchestrated process and cut approval times roughly sevenfold. Wellpointe automated 85% of complex medication orders, dropping manual intervention from one order in two to roughly one in seven, and scaled from 150 residents to more than a thousand without a back office to match. MySmartHeart runs 35,000 heart-monitoring cases a year across two countries on a single shared process, saving its team about two hours a day.

Different problems, same move: the process was re-engineered around governed AI, not patched with one more point tool. The recognition is catching up to the pattern. Gartner named Camunda a Visionary in its Magic Quadrant for Business Orchestration and Automation Technologies, and Forrester named it a Leader for Adaptive Process Orchestration in 2026, both pointing at the same strength, which is orchestrating the whole process rather than automating a corner of it.

Where to start

You do not re-engineer the enterprise all at once. Pick one journey. Choose the one costing you the most in time, in risk, or in patient and provider frustration: prior authorization, specialty-pharmacy onboarding, claims and appeals, or pharmacovigilance case intake. Map how it actually runs today, not how the diagram says it runs. Redesign it around the outcome you want, with agents on the volume and people on the judgment. Prove it on that one process, in weeks. Then carry what you learned, and the pieces you built, into the next one.

The organizations that lead healthcare and life sciences through the next decade will not be the ones that bought the most agents. They will be the ones that rebuilt their most important work to run on them, safely, and did it first. Your competitors are already choosing their first process. Choose yours.

See how the Great Re-Engineering applies to your operations → Healthcare & Life Sciences

Start the discussion at forum.camunda.io

Ready to get started?