For this blog, I’d like for you to picture a regional insurer whose claims’ team runs intake and adjudication on Pega. For over a decade, the workflow has accumulated layers of review: a claim arrives, a reviewer checks policy terms, and a decision gateway routes it to payment or denial.
Customer expectations haven't changed, but what has changed is how fast decisions are expected and how much of that review still requires a person. Every quarter that process stays frozen on a platform that's hard to extend is another quarter a competitor running on something more adaptable closes the gap on cycle time.
That gap between what the process does today and what AI tooling could improve is a compelling reason to consider moving beyond Pega. Earlier posts in this series, The Compounding Cost of Staying on Pega and Ready to Break Up With Pega?, covered the case for change and the migration choices. Here, we look at the conversion work itself: where AI helps, where it might fall short, and where human review remains essential.
Start with the converter—and know its limits
Camunda has had a Pega migration path for years. The original approach, still available today, is a free, open source converter in the migrate-to-camunda-tools GitHub repository. It’s a Maven project you run locally: point it at a Pega XML export and it generates a BPMN file that preserves the structure of the original process.
The converter has clear limits. Pega flows do not use BPMN as their native process standard, so a lossless translation isn’t possible. The output is a structural starting point, not a finished, executable process. Decision tables, connectors, or data mapping, and the output needs review in Camunda Modeler before it’s production-ready. The tool has the same limit every structural converter has: it can move shapes, but it can’t make judgment calls about what those shapes should mean in your environment.
What ProcessOS changes
Consider a fictional insurer with a decade-old Pega claims process. Discovery reveals several reviews that repeat checks. A redesigned workflow could remove the duplication, route low-risk claims through an agent, and retain human approval above a defined threshold. That is a different outcome from simply moving the existing process.
ProcessOS changes the work-around conversion. It uses documentation, system logs, and input from the people who run a process to understand its current state and inform a redesigned workflow. The aim is not to rewrite or relocate every existing step. It’s to determine where people, agents, and systems should each act.
That approach also shapes the migration engagement: short working sessions, frequent validation, subject-matter experts involved when expertise is needed, and forward-deployed engineers work in the customer environment.
Where AI-only conversion gets risky
AI-only conversion introduces risks that aren’t always obvious. Consider two errors that may not surface at deployment. A gateway condition can lose its FEEL prefix, so Camunda reads it as a fixed string instead of an expression, and the condition never evaluates. A call activity can miss a variable-propagation setting, which changes what data flows back to the parent with no warning.
AI can help with code and cleanup, but conversion quality depends on the model and the task. A deterministic converter produces repeatable output; AI-generated output requires close review. Neither approach can decide whether a business rule still makes sense, how eventual consistency affects a decision, or whether an audit requirement has been met. Those judgments belong to people who understand the process.
A Pega-to-BPMN example
Take a simple Pega request-review process with an approve/reject decision. Export the model to XML, then use Claude with Camunda’s BPMN modeling skill to generate a first-pass BPMN model.
The source process contains Pega-specific elements, names, and decision logic. The XML export provides the input for the conversion shown below.
The result is a BPMN model with a service task representing the subprocess, an exclusive gateway for the decision, branches labeled Approved and Rejected, and a converging gateway before the end event. It is a structural first pass, similar in purpose to the deterministic converter’s output, but generated through an AI-assisted workflow.
It is not production-ready. Task names are placeholders, and branch conditions still need to be connected to executable decision logic. More complex processes introduce additional subprocesses, branching, and layout challenges. Before deployment, teams must refine the model, configure its behavior, and validate each path against the original business requirements.
Practical migration sequence
A Pega-to-Camunda 8 migration follows five steps: assess, convert, clean up, re-engineer, and validate. Each step needs a checkpoint before work moves forward.
- Assess. Examine the actual Pega process—not just its documentation—and evaluate complexity before conversion.
- Convert the model. Use the Pega-to-BPMN converter or a validated discovery-to-model approach to establish a structural baseline. Review the output before proceeding.
- Use AI for code and cleanup. AI can assist with connector code, data mapping, and model refinement, subject to testing and review. It should not be assumed to produce executable, correct output on its own.
- Re-engineer rather than relocate. ProcessOS earns its name here: it uses what the process does today as context for what it should do now, including where an agent can reasonably take a step a person used to have to do manually.
- Validate the result. Review decision logic, data consistency assumptions, and audit requirements with process owners before go-live. Confirm that the implementation captures the records needed for the organization’s audit obligations.
Use AI where it helps. Keep people accountable for decisions that require business judgment. Agentic orchestration brings agents, people, and systems into a coordinated process, with explicit controls and visibility into how work gets done.
Ready to move beyond Pega? Explore Camunda's migration services and licensing incentives. Get your migration details.
