Fully declarative approach is the way to go for both IMHO. I'm more familiar with Kotlin than Kotlin. But its coroutines framework and structured concurrency are well aligned with something like this. Essentially you are doing is kind of a stateful form of structured concurrency where state is preserved in a DB and resilient against sub task failures, nodes dying, etc.

This reminds me of a product called restate. I talked to some people in that company a while ago. Their solution is built in Rust I think but they have clients for all sorts of platforms. Including Kotlin. Cool company and distributed workflow engines and agentic / long running workflows feel like a good match.

There are lots of other solutions in this space. I believe an ex Red Hat person is working on rebooting a workflow engine called Kogito based on something that orginally lived under their umbrella.

There's a long history of very enterprisy business process management stuff here. Lots of potential for overengineered solutions.

I once got sucked into a Spring Batch centric project and it was hopelessly overengineered for the requirements. Gave me a proper headache. Nothing was simple. Everything was littered in magic annotations causing all sorts of weird side effects. That's why I prefer declarative approaches with simple functions. Which is what the Kotlin syntax enables relative to Java. You can do the same technically in java but it quickly becomes an unreadable mess of function chaining.