L3 · Agent mesh
Twenty-six task agents. One output contract each.
Every agent declares what it produces, the gate it must pass, the Canon it writes, how it fails and how it is evaluated.
Task agents
26
Clusters
6
Domain agents
7
Orchestrator
Chief of Staff
Non-negotiable
The architectural rules the mesh runs under
Three constraints hold across every agent, workflow and market.
No agent-to-agent coordination. Every handoff goes through the Chief of Staff.
Swarms that self-coordinate cannot be audited, debugged or governed. The cost is a small latency penalty; the benefit is that every decision has a traceable path.
No output leaves the system without a Canon write-back.
This is what compounds. Without it, the platform is a very expensive drafting tool.
The autonomy ceiling is data, not code.
Legal must be able to ratchet down a ceiling at 23:00 on a Friday without a deployment.
“If everybody builds a messaging agent, and if everybody builds a… then you've got this proliferation of agents that don't talk to each other and actually have the opposite effect to the one that we're trying to drive, which is consistency and coherency.”
Supports · The most important sentence in the evidence base, and the reason this platform is designed the way it is.
Six clusters
How the register is organized
Clusters are contract families: sensing never publishes, production never decides, engagement never invents a position.
Everything the platform knows about the outside world enters here. These agents never publish; they write to the Signal Lake and the Stakeholder Graph.
The layer that turns signal into a recommended action with a projected consequence. These agents never publish either; they produce decisions and the evidence behind them.
The layer that writes. Every output is generated against the Narrative Canon and the Voice Corpus, and nothing leaves without a G5 RELEASE.
The layer that manages the two-way surface — inbox, journalists, regulators, employees. Relationships are owned by humans; these agents prepare and track, they do not own.
Not a separate system. The same agents under a different posture, plus three that exist only to run the incident spine.
The loop-closers. Without this cluster the platform gets no better after month one.
Appendix A · the register
Every task agent, in full
Filter by cluster, autonomy level or gate-bound output. Select a card for the full specification.
Showing 26 of 26 task agents
L3 · domain tier
Seven domain agents own outcomes, not tasks
Each domain agent holds a standing objective and commands a subset of the task register. Each has one human counterpart and a declared autonomy ceiling.
| Domain agent | Standing objective | Commands | Ceiling | Counterpart | Cost upside |
|---|---|---|---|---|---|
D-EXEC Executive & Strategic | Every principal is prepared, on-message and in voice for every appearance; the corporate narrative is coherent across all executive surfaces. | C2D3B5 | A2 | CCO | 12–16% task 8–11% |
D-MEDIA External & Media Relations | The company's earned-media position is known, defended and improving; every consequential journalist relationship has an owner and a current state. | A1A4C3D2D4 | A3 | VP Media Relations | 22–31% task 15–20% |
D-INT Internal & Employee | Every employee and franchise partner receives the corporate position in their language, in their context, within the cascade window — and their reaction is measured. | C4C6A2 | A3 | VP Internal Comms | 18–24% task 11–14% |
D-PA Public Affairs & Govt Relations | No regulatory or legislative development in any operating jurisdiction reaches the company as a surprise; positions are filed on time and consistent across jurisdictions. | A3B1C3 | A3 | SVP Government Relations | 16–22% task 11–15% |
D-ESG Sustainability & ESG | Every sustainability claim is substantiated, in-scope and current; progress is reported without overstatement; greenwash exposure is monitored continuously. | A6B4C5 | A3 | Chief Sustainability Officer | 21–30% task 14–18% |
D-DIG Digital & Multimedia | Owned channels carry the current narrative; content is produced at market scale without narrative divergence; the company's representation in AI answer surfaces is monitored and contested. | C5C6A5 | A3 | VP Digital | 19–27% task 14–18% |
D-OPS Ops, Planning & Analytics | The function knows what it did, what it cost and what it changed; reporting is a query, not a project; every workflow’s autonomy and gate load is tuned. | F1F2D1 | A4 | Head of Comms Operations | 28–39% task 21–27% |
Why there is no crisis domain agent
Crisis is deliberately not a domain agent. It is a posture that re-tasks all seven. A standalone crisis team that spins up on declaration is the current model, and it is why the first ninety minutes are always lost. In CHORUS, crisis posture reconfigures the standing organization: ceilings drop, gate deadlines compress from hours to minutes, the E-cluster activates, and D-EXEC and D-MEDIA are placed under a single incident commander.
Point of attack
What each domain agent is pointed at first
Executive & Strategic
Deliberately capped. C2 drafts; the human owns voice and position.
Productivity
27–37%
Cost
12–16%
External & Media Relations
A1 Media Monitor; D2 Journalist Targeting (pitch stays human).
Productivity
33–48%
Cost
22–31%
Internal & Employee
C4 Internal & Franchisee Cascade — the largest single win at McDonald's.
Productivity
34–47%
Cost
18–24%
Public Affairs & Govt Relations
A3 Policy & Regulatory Radar across 114 jurisdictions.
Productivity
29–41%
Cost
16–22%
Sustainability & ESG
Evidence Ledger; claim substantiation; B4 Narrative Health.
Productivity
34–47%
Cost
21–30%
Digital & Multimedia
C5 Social Content; C6 Localization at market scale.
Productivity
38–53%
Cost
19–27%
Ops, Planning & Analytics
F1 Performance & Attribution; automated reporting from the Evidence Ledger.
Productivity
37–52%
Cost
28–39%