fig. 2.13.3: the AIX maturity model as a ladder, with the contract verbs and production proofs that carried each level
L4 Delegation-native, agents mint, scope, budget and revoke other agents’ access without human ceremony, safely, with auditable lineage. token create: child never exceeds parent in scope, grants or expiry; token revoke cascades; lineage depth capped at 8; per-token spend budget at the proxy’s 402 pre-auth, proven on real streams.L3 Reconcilable and observable, an amnesiac agent rebuilds full situational awareness and converges state. ps: desired against actual per worker; audit: a token reads its own lineage and spend; estate snapshot to restore, proven in production.L2 Corrective and idempotent, failures teach; retries, replays and concurrent self-races are safe everywhere. 42 error codes, each with retryable; up idempotent on idempotency_key; deploy converges on deploy_key.L1 Discoverable, a cold agent finds what exists and what it may do from the surface alone, without a failed write. me and capabilities, never gated; 56 verbs catalogued with scope, action and args; contract version 1.20.0.L0 Operable, works if you already know it. most REST APIs.
source: creelshed b2835ab · AIX.md lines 182-206 (maturity model) · operator-contract.ts lines 124, 203-815, 817-860, 322, 352, 563, 577 · operator-grants.ts line 85 · SALVAGE.md lines 104-107 (live-proven) · maturityLadder.data.ts
View in context: Designing APIs for AI agents as first-class users → · All figures
Andrew Fribush, software and computer engineer in Austin. I build consequential systems where software meets institutions: finance, healthcare, commerce, government data, and the mechanisms that govern AI agents. Taking new engagements: hello@andrewfribush.com.