Blueprint (deterministic and agentic steps)
A workflow defined in code that mixes fixed steps, such as lint, build and git, with steps where an agent decides what to do.
Stripe describes blueprints in its Minions posts: workflows defined in code that direct an unattended coding agent run, combining “the determinism of workflows with agents’ flexibility”. Checking out a branch, running the linter, pushing and opening a pull request are plain code; understanding the task and editing files is an agent node. It sits between a fixed agentic workflow and a free agent.
Deterministic steps are cheaper, faster and never forget, so anything that can be scripted should be. The agent gets fewer decisions and less room to go wrong, and failures are easier to trace to a node. Blueprints usually run as background agents or in headless mode in CI; a dynamic workflow is the variant where the model writes the script itself.
Sources
Related terms
- Agentic workflow (workflow vs agent): A process where your code fixes the steps and a model does some of them, as opposed to an agent that decides its own steps.
- Dynamic workflow: An orchestration script the agent writes at run time. A runtime executes it, starting many subagents in phases and combining their results in code.
- Background coding agent: A coding agent that runs in the cloud on its own, takes an issue or a task and comes back later with a pull request for review.
- Validation loop (agent backpressure): Commands a coding agent runs after every change (tests, types, lint, build) and fixes against until they pass, so it checks itself before a person looks.
- Headless mode (coding agent in scripts and CI): Running a coding agent without an interactive session, from a script or pipeline, with the prompt, allowed tools and output format given as arguments.
