Agentic Workflows in Scala (Without the Buzzwords)

"In the last few months, agentic workflows have been showing up everywhere - especially in discussions about large language models and autonomous systems. Depending on who you ask, the term can mean anything from "LLMs calling tools" to fully autonomous agents running in opaque loops. But here's the uncomfortable truth: most of what's being described isn't new. We've been building these kinds of systems for years - just without the buzzwords."

"What is new is the framing - and that framing matters. By centring the conversation on autonomous agents and hidden reasoning loops, many teams are being nudged toward implicit control flow and opaque decision-making. But the systems that actually hold up in production tend to need the opposite: explicit state, clear decision points, and durable workflows that can pause, resume, and evolve over time."

Agentic workflows are long-running workflows that maintain explicit state, make decisions from intermediate results, and orchestrate tools across time. Emphasizing explicit state and clear decision points prevents opaque control flow and supports durability, pause, resume, and evolution. Systems that succeed in production favor explicit orchestration and durable workflow state over implicit, hidden agent loops. Implementing agentic workflows requires explicit decision-driven designs, persistent state models, and transparent tool integration. Scala, Workflows4s, and Apache Pekko provide primitives for building such systems with durable execution, explicit state management, and safe interaction with external systems.

Read at Medium