Workflow Engine
A system that runs a defined sequence of steps reliably, handling ordering, retries, and failures.
Think of it like
An assembly line — each station does its part in order, and if one jams, the line knows how to respond.
Example
A workflow engine runs “extract → validate → summarize → email” for every uploaded document, retrying any step that fails.
How it actually works
Workflow engines predate LLMs — they’re the durable backbone for multi-step processes, guaranteeing steps run in order, survive crashes, and resume where they left off. For AI systems they’re increasingly paired with agents: the engine handles the reliable plumbing while the model handles the judgment. The tradeoff is rigidity — engines love predictable flows and chafe against the open-ended improvisation agents are good at.
For product teams
Use an engine when a process must run reliably every time; use an agent when it needs to improvise.
For engineers
A durable executor that sequences steps with ordering, retry, and crash-recovery guarantees.
Related
- Deterministic Workflow — A fixed flow it typically runs.
- Dynamic Workflow — The flexible counterpart it can also drive.
- Agent Orchestrator — The LLM-driven coordinator it contrasts with.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome