Notebook · Business as an information system
Making a business
legible to AI.
My view of a business as a system of roles, information, and decisions, and what that implies for AI.
Part one · The mental models
Eight mental models I keep returning to.
Mental model 08 · The centerpiece
I think useful AI starts with making the organization legible.
First, become legible
Before you can automate a business, you must understand it. Before you understand it, you must model it: what each function does, consumes, produces, and decides, and where judgment is required.
Then, the design questions become clearer
An explicit model of the organization helps me identify candidate uses for copilots, automation, and prediction. It also reveals where context is missing or human judgment must remain. Those are hypotheses to test, not guarantees of safe automation.
Mental model 02, in depth
The layered enterprise.
Start at the top and work your way down. Executives design at the highest level of abstraction; intelligence lives at the core.
From purpose to a specific decision
- StrategyChoose the objective and acceptable tradeoffs.
- Functions and workflowsDefine the capabilities and how work moves.
- Tasks and informationLocate decisions and the evidence they require.
- IntelligenceApply context and reasoning where they can improve those decisions.
The map · Surfaces of intelligence
Two surfaces, one system.
Intelligence improves how the company operates, and what the company sells. The same discipline, applied inward and outward.
Enterprise Intelligence: how the business operates
I can imagine a different decision context for each function:
Finance: forecasting, margin optimization, anomaly detection · Sales: qualification, account planning, renewal prediction · Customer Success: churn prediction, health scoring · Support: routing, copilots, predictive escalation · Product: prioritization, telemetry, roadmap simulation · Engineering: copilots, review, incident investigation · Operations: scheduling, capacity, exception detection · HR · Legal · Data: search, contract review, data quality.
Product Intelligence: what the company sells
From telling customers what happened to telling them what to do:
Predictive maintenance: “These assets will likely need service within their next operating window.” · Utilization: “Reallocate this resource to avoid renting another.” · Efficiency: “You're spending more than comparable operations due to idle behavior.” · Safety: “This pattern correlates with elevated risk.”
I’m interested in interfaces that bring relevant context to a decision, rather than adding another place to search for it.
The wiring · Role Programming Interface
The real-time enterprise.
I imagine a system organized around roles: each has its own ThoughtSpace and exposes a bounded set of capabilities. A Role Programming Interface is my model for how those capabilities could communicate through explicit contracts.
The translation tax
Today, one department needs something from another and a human relays it, losing fidelity and time at every hop.
Exposed functions
Each department publishes callable functions: request_quote, validate_budget, flag_churn_risk. The org becomes an API.
Real-time at the edge
Shared contracts can connect capabilities without requiring a single database. Access checks, routing, and audit records belong in the system surrounding the model.
Sales exposes account state and expansion signals. Finance exposes budget validation and discount approval. Customer Success exposes account health. Product exposes usage telemetry. The RPI bus routes the call, enforces the contract, governs the data, and writes the audit log.
A governed capability handoff
- SalesSends account context and an expansion request.
- Shared boundaryAuthenticates the caller, validates the contract, routes the request, and records it.
- FinanceChecks pricing and authority; escalates exceptions.
- Customer successReceives the approved result and the context needed to act.
The possibility that interests me is less time spent reconstructing context, and more attention available for judgment.