Dynamic Workflow
A process where an agent decides the next step at runtime instead of following a fixed path.
Think of it like
Improvising a road trip — you pick each turn based on what you find, rather than following a printed itinerary.
Example
Given a research question, the agent chooses whether to search, read, calculate, or ask a follow-up, and the path differs every time.
How it actually works
Dynamic workflows trade predictability for flexibility: the model chooses each step, so the system can handle situations no one scripted. That’s exactly what makes agents powerful and exactly what makes them hard to test, budget, and trust. The mature pattern is a blend — a deterministic skeleton for the parts you understand, with dynamic decision-making reserved for the genuinely open-ended steps.
For product teams
Dynamic flows handle the unexpected but cost you predictability — reserve them for where flexibility truly earns its keep.
For engineers
The model selects each next action at runtime, so the execution path is decided dynamically rather than predefined.
Related
- Deterministic Workflow — The fixed-path opposite.
- Agent Loop — The loop that makes it dynamic.
- Agent Orchestrator — The controller often driving it.
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