Composable AI:
Build Prod, Not God
OUR MISSION
Intelligence becomes useful when people can build dependable software around it.
THE COMPOSITION PROBLEM
Software grows by combining smaller parts. Each part has an input, an output and a contract that other parts can rely on.
The manifesto argues for bringing that discipline to AI: a bounded task, a typed answer and a clear place for uncertainty.
A NEW BUILDING BLOCK
A conversational interface asks a person to interpret the result. A machine-oriented interface can make a smaller decision inside an existing program.
That leaves routing, retries, permissions and side effects in ordinary code, where the application can test and control them.
The narrow component can then be reused across many different workflows.
DEPENDABLE COMPOSITION
The guide's implementation mirrors that idea: explicit local contracts, visible uncertainty and separate behavior for observed source states and proposed fixtures.
Components can be inspected on their own, then assembled into a complete page without hidden dependencies.
THREE DESIGN CONSEQUENCES
- Make the output type visible.
- Show the boundary of each action.
- Give the user a way to recover.