Operating Model: Mission Control for AI

Operating model, not pilots

Executive framing

• “AI delivery” fails when it’s treated as a tool rollout.
• It succeeds when it’s treated as an operating model change:
  • decision rights,
  • escalation,
  • accountability,
  • and practice under failure.

What changes in the agent era

• Systems don’t just predict — they can act (via tools, workflows, and delegation).
• That increases speed and surface area.
• The question becomes: what is allowed to run unattended, and how do we stop it?

Organisation ↔︎ architecture

Institutional Character

Bezos memo to Amazon in 2002

The API Mandate

• All teams will henceforth expose their data and functionality through service interfaces.
• Teams must communicate with each other through these interfaces.

• There will be no other form of inter-process communication allowed: no direct linking, no direct reads of another team’s data store, no shared-memory model, no back-doors whatsoever. The only communication allowed is via service interface calls over the network.

• It doesn’t matter what technology they use.
• All service interfaces, without exception, must be designed from the ground up to be externalizable. That is to say, the team must plan and design to be able to expose the interface to developers in the outside world. No exceptions.

Duality of Corporation and Information

• What is less written about is corporate structure.
• This information infrastructure is reflected in the corporation.
• Two pizza teams with devolved autonomy.
• Bound together through corporate culture.

Conway’s Law

Any organization that designs a system (defined broadly) will produce a design whose structure is a copy of the organization’s communication structure.

Conway (n.d.)

Amazon and Agility

• Company prides itself on agility.
• Operates through a system of devolved autonomy
  • Teams have defined inputs and outputs.
• Corporate culture bonds them togethether.
  • Customer obsession
  • Ownership

Cambridge

• Over 800 years old.
  • Less commonly thought of as agile.
  • Reliable institutional character.
  • Amazon in comparison is more mecurial.

Organisational Data Science

• Claim:
  • For data driven decision making to work there must be an interplay between institutional character and the information infrastructure.

Cultural Transplant

• Claim:
  • Transplanting another organisation’s decision making infrastructure directly leads cultural rejection from the new host.

Mission control: roles, escalation, practice

Mission control primitives (minimum viable)

• Roles: operator, reviewer, owner, risk/compliance, incident lead.
• Decision rights: automate / assist / escalate (make it explicit).
• Runbooks: “what do we do when it fails?” (before it fails).
• Drills: pre‑mortems, simulations, and incident exercises.

Escalation design

• Define triggers for human escalation (uncertainty, novelty, drift, complaints, anomalies).
• Include “pause when unsure” as a first-class action.
• Make override real: time-bounded decisions, rollback paths, and audit trails.

How to structure a rollout

• Sandbox → supervised operation → constrained autonomy → broader autonomy.
• Increase autonomy only when monitoring + incident response prove reliable.
• Treat each step as a control upgrade, not just a capability upgrade.

What to do on Monday (practical)

• Pick one automated (or candidate) decision and document:
  • owner/accountable person,
  • decision rights (what can run without approval),
  • triggers + escalation path,
  • logs/audit trail expectations,
  • “kill switch” and rollback.