L3 · Execution
Watch the work move and the outputs appear.
Pick a line and press play. Each step names the agent that holds it, its autonomy level, the gate that can stop it and the artifact it produces. Gates halt the run until a human releases them.
Steps simulated
98
Handoffs
91
Artifact classes
12
Gate stops
15
Choose the situation
The situation
An allegation involving a named individual surfaces. Everything downstream is privileged, evidentiary and irreversible if it is wrong.
Trigger
Allegation received through any channel
Clock
Deliberately slow — ceiling held at A2 by the General Counsel
Ceiling rationale
A2. Deliberately the lowest ceiling in the platform. Under crisis posture the value of speed is real but the cost of a wrong autonomous action is unbounded, so the ceiling drops and the gate deadlines compress instead.
Run state
Steps done
0/18
Artifacts
0
Gates cleared
0
Human touches
0
Canon stores written
0
Overrides
0
Autonomous so far
—
Detect early signals — tribunal filings, employee-channel sentiment inflection, union communications, journalist inquiry patterns, advocacy-group activity
Fuse into a single early-warning card with the coalition forming around it and the closest precedents with their outcomes
Map exposure — markets, entities, franchise versus company-operated, legal jurisdictions, regulatory interest, prior related matters
ESCALATE — CCO, GC and CHRO convene. The platform hands over a situation, not a plan.
Assemble the factual record from company systems — what is known, what is contested, what is under investigation, what is subject to legal privilege
Retrieve precedent — how comparable matters resolved at peers, what responses helped, what responses made it worse
G1 POSITION — the company's stance. This is a leadership judgment about values, not a communications choice, and it is made by people.
Draft the position statement with the human owning structure and substance; agent supplies precedent, constraint checks and alternatives
Simulate reaction — affected workers, employee base, union leadership, the reporting journalist, the regulator, franchise operators, investors
Red-team hard — the hostile headline, the three weakest claims, the sentence that will be quoted against the company for years
G3 RISK — litigation exposure, regulatory exposure, employee-trust exposure, and the reading that the least charitable reader will take
G4 VOICE — any first-person executive content. The principal decides, not a comms generalist.
Produce the supporting set — employee communication, franchise-operator briefing, press Q&A including the questions nobody wants, regulator materials
G5 RELEASE — full stop, no batching, no one-click. Every artifact individually authorized.
Monitor reception across every affected cohort, especially employee and operator sentiment
Track every commitment made publicly — investigations promised, policies changed, timelines given — and chase them to completion
Maintain the long tail — this runs for years. Track anniversary coverage, related filings, and every subsequent inquiry against the established position.
Write back — what was said, what was promised, whether it was delivered. The commitment ledger is the reputational asset here.
Output shelf
0Everything the run has actually produced. Click an item to read the output contract the producing agent is held to.
Nothing produced yet. Run the situation.
Canon touched
No write-backs yet.
Architectural rule R2: no output leaves the system without a Canon write-back. That is the difference between a tool that produces content and an organization that gets smarter.
Output taxonomy
What a step actually produces
Every specified step leaves an output. Grouping them into classes shows what was produced, by whom, under which level, and what it changed.
The work reached the audience it was built for.
Publishable language in the house voice.
The organization is permanently smarter for it.
A judgment with its drivers attached.
What the company says, and why it is defensible.
The same position, landed in-market.
A position of record with an external body.
A claim is substantiated and traceable.
What it changed, counted rather than claimed.
A named human accepted the consequence.
Something in the world was noticed and named.
A human is prepared to act or answer.
Why the bench behaves this way
Three rules govern every run
The simulation obeys the same three architectural constraints the platform is specified under.
R1
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.
R2
No output leaves the system without a Canon write-back.
This is what compounds. Without it, the platform is a very expensive drafting tool.
R3
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.