What it is

The architecture that emerges across the evidence is a discovery-to-governance lifecycle: a user expresses intent; the runtime retrieves current capability descriptions; the model selects tools and binds arguments against formal schemas; an authorisation layer establishes who or what is acting, for whom and with which limits; an execution substrate handles retries, timeouts and compensation; and an evidence layer records provenance, traces, policy decisions and consents.

Why it matters

API-calling is only one layer of an agent-capable stack. Reliable deployment depends on capabilities outside the model: contracts, discovery, identity, execution, policy and observability/provenance. None substitutes for the others.

Relates to: Agent-ready, not agent-trusting, The federation layer.

Terms

Glossary terms used here: Compensation, Provenance.