AgentRun turns ad-hoc agent scripts into a DSL

Most AI agent work still starts life as a script. Someone connects a model to a few tools, adds prompts and conditions, then adjusts the code whenever the process changes. That’s fine for testing, but it gets messy when a useful script becomes part of how a team actually operates.

AgentRun is a small Show HN project that tackles that problem by turning scripted agent behaviour into repeatable, declarative workflow definitions. In plain terms, instead of relying on a one-off script to describe what an agent should do, the workflow is defined in a format intended to be clearer and more reusable. The project sits in the growing workflow-as-code pattern, where business processes are treated more like managed system components than loose automation experiments. It has attracted 45 points and 10 comments so far. You can see the project details in the original repository.

For operators, the important question isn’t whether this particular project becomes widely used. It’s the direction it points to. Agent workflows need to be repeatable, inspectable, and easier to change without rebuilding everything from scratch. If an AI process handles lead research, document checks, customer routing, or internal reporting, you need to know what it is meant to do and where it fits into the wider operation.

This is the kind of thing we build into an AI command centre, where workflows are visible, structured, and connected to real operating processes rather than buried in isolated scripts.