---
name: agentic-enterprise-architecture
description: "The cross-layer reference for running AI agents inside an
  enterprise: the seven capabilities an agent platform provides across fourteen
  layers, the two rules (policy enforced outside the model; the audit trail
  written where the agent cannot touch it), the four deterministic zones (models
  may take permission away, never grant it, and inside the zones may not approve
  at all), the delegation chain, enforcement outside the model, the
  data-to-memory pipeline, the learning flywheel, the autonomy contract (A0 to
  A5 by L0 to L3), the cross-cutting concern matrix, and a plain-words glossary.
  Use whenever a question concerns how an enterprise should architect, govern,
  secure or supervise agents, or when a term like A3, ID2, the audit trail, the
  oversight gate or a deterministic zone needs its definition."
license: CC-BY-SA-4.0
compatibility: Reference material only. No tools, network access, or scripts
  required; any agent that can read Markdown can use it.
metadata:
  track: enterprise
  kind: reference
  version: 2026.10.15
  source: https://www.agenticarchitectureskills.com/skills
  bundled: "14"
---

# The agentic enterprise, cross-layer

Twelve pages, ordered so that reading them in sequence builds the argument. Load only what the question needs.

| File | What it answers |
|---|---|
| `references/01-target-state-architecture.md` | The whole design on one page: seven capabilities, the two rules that bind them, your own agents and vendor-embedded agents, three builds by size and regulatory intensity |
| `references/02-plain-words-glossary.md` | Every term the other pages use, in plain words |
| `references/03-deterministic-zones.md` | The rule that models may take permission away, never grant it; the four places a model may not approve at all; and how to build each boundary |
| `references/04-identity-and-delegation.md` | How an action stays traceable to the person who asked, hop by hop, with the standards |
| `references/05-enforcement.md` | Why the rules that stop an agent live outside the model, and the three enforcement points that hold them |
| `references/06-data-to-memory.md` | How company information reaches an agent and what it may remember |
| `references/07-learning-flywheel.md` | How behaviour changes without anyone losing control: evals, promotion gates, demotion |
| `references/08-autonomy-contract.md` | How much an agent may do alone, decided per task, with the controls each level requires |
| `references/09-concern-matrix.md` | Ten cross-cutting concerns against fourteen layers: who owns and who enforces each |
| `references/10-master-target-state.md` | The three complete architectures, keyed to size and regulatory intensity |
| `references/11-maturity-model.md` | The autonomy and learning axes in full, with the gates between levels |
| `references/12-vision.md` | What an agentic enterprise is, and the shape of the residual human work |

## Finding things quickly

The files are long. Search rather than read end to end:

- A capability's responsibilities: `grep -n "capabilit" references/01-target-state-architecture.md`
- A specific autonomy level: `grep -n "A3\|A4" references/08-autonomy-contract.md references/11-maturity-model.md`
- Whether something is a zone: `grep -n "zone" references/03-deterministic-zones.md`
- Who enforces a concern: `grep -n "C[0-9]" references/09-concern-matrix.md`
- A term: `grep -in "<term>" references/02-plain-words-glossary.md`

## How to use it

Two rules travel with every answer drawn from this material.

**Policy is enforced outside the model.** An instruction in a prompt is a preference; the same rule becomes a control when a gateway or policy decision point evaluates it.

**The audit trail is written where the agent cannot touch it.** Anything an agent reports about itself does not count as evidence.

## What this reference refuses to say

It publishes no timeline, no human-to-agent supervision ratio, and no build-versus-buy figures, because the evidence for those does not exist in a citable form. Six cross-cutting concerns have no complete published answer anywhere and are marked as open rather than papered over. Products are named for orientation as of the date on the wall chart, are representative rather than exhaustive, and are not endorsements. Where a number comes from a vendor, the text says so.


## Applying this guidance

Treat the supplied references as a starting point, not proof that a recommendation fits the task. Check important claims against dated external evidence. Distinguish proposed industrial examples from delivered work and author-reported experiments. Use the user's actual industry and constraints; utilities and mining are examples, not universal requirements. Keep working-file and internal repository references out of reader-facing reports.

## Additional task references

- [What an agent platform provides](references/architecture-system-view.md): consult when this aspect is part of the task.
- [Follow one inspection task](references/architecture-task-walkthrough.md): consult when this aspect is part of the task.