Workflow vs agent
In a workflow, LLM calls follow a path your code decides. In an agent, the model decides its own next step and which tools to use, in a loop.
LangGraph’s documentation puts it plainly: “Workflows have predetermined code paths and are designed to operate in a certain order,” while “Agents are dynamic and define their own processes and tool usage.”
Frameworks now ship five workflow patterns as building blocks: prompt chaining, routing, parallelisation, orchestrator–worker and evaluator–optimiser. Many systems marketed as agents are one of these.
Practitioners lean predictable. In Stack Overflow’s 2026 pulse survey, 69% of developers run a single agent and 68% prefer predictable single-agent setups to complex multi-agent ones.
If it calls an LLM, it’s an agent.
If code decides the path, it’s a workflow. Often the better choice.
Draw the control flow before you pick a framework. Mark every step where the model, not your code, chooses what happens next.
Ask vendors where the model decides, and how they test those points.
Sources
- LangGraph docs, Workflows and agents · accessed Oct 2026
- Stack Overflow, Agents on a leash (pulse survey, n = 1,100) · 27 May 2026