04 · AI Operating Models
The operating model is part of the AI product.
When AI systems participate in real workflows, technology decisions also become decisions about authority, accountability, risk, and work.
Define decision rights
Be explicit about what an agent decides, what requires approval, and who owns the outcome.
Design evaluation into the system
Define how you measure quality before increasing autonomy.
Make behavior observable
Observe more than infrastructure. Capture the signals needed to understand actions, decisions, and failures.
Treat memory and context as architecture
Decide what the system needs to remember, for how long, with what provenance, and under whose control.
Design the human path
Human oversight needs clear triggers, useful context, and defined responsibility.
Think about failure before autonomy
Ask what happens when the model is wrong, the context is incomplete, a tool fails, or a policy changes.