StewAI Blog · AI Workflow Design · 12 min read
Do You Know How the Work Is Done? A 2x2 for Recipes vs Agents
Two questions decide whether to use an autonomous agent or a deterministic recipe: can you specify how the work is done, and will you run it again. A 2x2 for choosing, and the arrow that turns repeated exploration into reusable infrastructure.
Frequently asked questions
Should I use an autonomous agent or a deterministic recipe?
Answer two questions. Can you specify how the work is done, and will you run it again? Known, recurring work belongs in a recipe that is typed, versioned, and auditable. Unknown, one-off work suits a bounded autonomous agent. Unknown but recurring work starts agentic and is then crystallized into a recipe.
What are the four quadrants in the recipes versus agents framework?
Known and recurring is the operating asset, encoded as a recipe. Unknown and recurring is the frontier process: explore with an agent, then crystallize. Unknown and one-off is a bounded agentic run. Known and one-off is a disposable execution you should not over-engineer.
Why do structured workflows beat raw autonomy in production?
End-to-end success is the product of the per-step success rates, so errors compound. At 85 percent reliability per step, a 10-step task succeeds only about one time in five. Keeping chains short and bounded, with deterministic logic between probabilistic steps, is how production systems stay reliable.
Read the full article on StewAI