Label-Driven Agentic Workflows: Building Autonomous Software Pipelines Without a Workflow Engine

Short summary

The article proposes using issue tracker labels (GitHub, GitLab, Jira) as a distributed state machine for chaining AI agents into autonomous software pipelines. Each agent polls for its assigned label, performs its task, and swaps labels to hand off to the next agent—eliminating the need for a separate orchestrator or message queue. The approach leverages existing audit trails and supports human-in-the-loop interventions via simple label changes.

• •Use issue tracker labels as workflow states for AI agent pipelines
• •Agents poll for their label, work, then swap labels to hand off
• •No separate orchestrator, database, or message queue needed

Generated with AI, which can make mistakes.