Workflow Patterns for AI Agentss
Workflow patterns are reusable ways to compose several model calls into a dependable system, such as prompt chaining, routing, parallelisation, orchestrator-workers and evaluator-optimiser loops. They sit between a single prompt and a free-running agent, giving most of the benefit with far more control over cost and behaviour.
Stage: Patterns · Part of the AI Agents roadmap · Last reviewed 19 November 2026
What to learn in Workflow Patterns
- Chain two prompts with a validation gate between them
- Route an input to a specialised prompt based on its category
- Run independent subtasks in parallel and merge their results
- Build an evaluator loop that retries until a quality check passes
- Choose the simplest pattern that solves a stated problem
Workflow Patterns interview questions
What is prompt chaining and when does it help?
Prompt chaining splits a task into sequential model calls where each output feeds the next, often with a programmatic check in between. It helps when one big prompt is unreliable, because each smaller step is easier to get right, test and debug on its own.
How does the evaluator-optimiser pattern work?
One model call produces a draft and a second call critiques it against clear criteria, then the first revises. The loop stops when the critique passes or after a fixed number of rounds. It works best when you can state quality criteria concretely.
Where Workflow Patterns fits in the AI Agents roadmap
Workflow Patterns belongs to the Patterns stage (advanced level), which takes about 18 hours. Architectures.
Other units in this stage
Back to the full AI Agents roadmap