Supporting page

API Components

API is easiest to apply when autonomy is decomposed into concrete, inspectable capabilities. This page lists the component dimensions commonly used to explain why one process is “more autonomous” than another.

1) Goal Interface

• Accepts a goal plus constraints (scope, time, cost, policy) and a success definition.
• Rejects ambiguous goals or requests clarification rather than silently guessing.
• Produces a structured “contract” the rest of the system can reference.

2) Planning & Re‑planning

• Creates an executable plan with checkpoints.
• Updates the plan when dependencies fail, inputs change, or new information appears.
• Maintains traceability: what changed, why it changed, and what the new risks are.

3) Action Execution

• Can operate tools (APIs, CLIs, UI automations) and manage state.
• Uses idempotent operations where possible; avoids repeating unsafe actions.
• Knows when it lacks permission and how to request it.

4) Observation & Verification

• Checks outcomes against the success definition.
• Distinguishes “completed task” from “completed steps.”
• Measures quality (not just status) using tests, validations, or sampling.

5) Exception Handling

• Recovers from common failures (timeouts, missing data, transient errors).
• Can pause safely, roll back, or reroute the plan.
• Escalates with a clear summary: what happened, what it tried, and what it needs.

6) Memory & Context

• Maintains short‑term working memory for the current run.
• Optionally learns long‑term patterns (with controls) to reduce repeated failures.
• Separates durable facts from transient observations.

7) Governance & Controls

• Explicit permissions, audit logs, and approval gates for high‑impact actions.
• Rate limits, safety policies, and deterministic fallbacks.
• Clear accountability: humans remain responsible for outcomes.

These components can be scored individually, then combined into an overall API level. See Methodology.