The Autonomous Enterprise Stack

2026-05-07
Synthesis
Governance Runtime Foundation

Six layers. No vendors. One operating model.

Every conversation in this session collapses into this stack. Top three layers are governance — they decide what is allowed. Middle two are runtime — they enforce it. Bottom layer is foundation — it makes everything else possible.

Layer

06

Outcome Governance

The KPI panel, the agent owners, the review cadence. The layer your board sees.

Owned by

Executive sponsor + agent owners

Layer

05

Policy & Authority

Blast-radius zones, hand-off postures, role definitions, accountability chain. The contract between humans and agents.

Owned by

Risk + business + architecture

Layer

04

Trust Controls

Pre/in/post-flight validators. Kill switches at all four levels. Replayable audit trails.

Owned by

Platform engineering + security

Layer

03

Agent Runtime

The orchestration topology, planner, memory, reflection loop. Where the agent actually executes.

Owned by

AI platform team

Layer

02

Tools & Integrations

The catalog of actions agents can take — APIs, databases, applications — with idempotency, scope, and rate limits enforced at the edge.

Owned by

Application teams

Layer

01

Identity & Foundation

Non-human identity, scoped credentials, observability, network policy. The substrate; if this is wrong, nothing above it can be right.

Owned by

Identity + infrastructure

How to read this stack: most enterprises have layers 1 and 2 (some better than others). Most are figuring out layer 3. Few have layer 4. Almost none have 5 and 6 written down. The maturity of an agentic enterprise is measured top-down, not bottom-up.