Fleet 1.13:Teams are now shipping 5x more PRs with autonomous pipelines.See what's new →
FleetFleet
Glossary

Agent Pipeline

An agent pipeline is a defined sequence of stages in a software delivery workflow, where each stage is executed by one or more AI agents and the output of each stage is the input to the next.

Pipelines provide structure for complex, multi-step workflows that would otherwise require ad-hoc coordination. A typical coding pipeline might include: requirements analysis, implementation, testing, code review, security scan, and deployment. Each stage has a defined owner (which agent or human role executes it), defined inputs and outputs, and defined completion criteria.

Pipelines can be synchronous (each stage must complete before the next begins) or allow parallel stages where independent work can proceed simultaneously. Approval gates can be inserted between any stages to introduce human checkpoints.

The term 'pipeline' is sometimes used loosely to describe any multi-agent workflow. The more precise usage reserves it for pre-defined, stage-based flows — distinct from reactive event-driven chains where the sequence is determined at runtime by which events are published and which subscriptions match.

How this relates to Fleet

Fleet's codebase includes pipeline infrastructure, though the preferred coordination model for autonomous agent fleets is the reactive event chain rather than pre-defined pipelines. The fleet pipeline CLI commands support running and monitoring pipeline-based workflows where a fixed stage sequence is appropriate, such as structured release processes that require explicit approval at each stage.

Frequently asked questions

When should I use a pipeline versus a reactive chain?

Use a pipeline when the sequence of steps is fixed and known in advance, the completion criteria for each step are clear, and auditability of each stage transition matters. Use a reactive chain when the workflow is open-ended, different tasks may require different step sequences, or you want agents to self-organize around events rather than follow a predetermined script.

How are failures handled in an agent pipeline?

Each stage should have an explicit failure handler: retry with the same agent, escalate to a human reviewer, roll back to the previous stage, or halt the pipeline. Unhandled failures that silently propagate to the next stage are a common source of hard-to-diagnose issues in pipeline-based systems.

Run your first agent fleet

One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.