The cost of the wrong question
A fast answer to the wrong question can be more dangerous than a slow answer to the right one. Analysis begins by identifying the decision that matters.
Field notes.
Small observations about building durable businesses and intelligent systems. Short notes, sharper questions.
A notebook of working principles001—020
A metric is a contract
The number is the final artifact. The contract behind it includes population, time, exclusions, ownership and a shared business definition.
Portability is respect
The freedom to leave a service is part of why customers can trust it. Architecture should make departure possible before it is requested.
Permission is not ownership
Reading a dataset, enriching it, deriving a model and training on it are different rights. A single checkbox cannot resolve the distinctions.
The hidden price of convenience
Fast integration may reduce today's engineering work and increase tomorrow's bargaining cost. Measure both before calling a choice efficient.
AI needs a business dictionary
A language model may read SQL perfectly and understand the business incorrectly. Defined terms and approved metrics are critical infrastructure.
No proof, no success
A provider's HTTP 200 can mean only that a request was accepted. Reliable systems distinguish accepted, delivered and verified outcomes.
Every shortcut has a future
A shortcut is not always wrong. It becomes debt when its consequences are invisible, unmanaged or falsely described as the finished architecture.
Shared, not entangled
Common services should make products stronger, not force them into a single release cycle. Reuse without clear ownership can become coupling.
A model is not a policy
Prediction identifies a possible future. Policy determines who may act, with which safeguards and under whose responsibility.
Make uncertainty visible
Confidence should not come from fluent language. Good interfaces reveal what is known, what is assumed and what evidence is missing.
Architecture is economic judgment
The choice to centralize, abstract, automate or remain simple is a business decision with future costs, not merely a technical preference.
A system must remember why
An audit trail that records only what happened is incomplete. Durable systems should preserve why an action was authorized and which rule applied.
The art of reversible decisions
Move quickly on actions you can reverse and deliberately on those you cannot. Good product design makes the difference explicit.
Humans should keep the last word
Automation can compress work but should not erase accountability. Sensitive decisions need review, explanation and a clear path to correction.
Build one useful thing completely
An unfinished shared platform can delay every product depending on it. A narrow, stable capability is often more valuable than a broad promise.
A platform's real customer is its products
Shared infrastructure succeeds when adopting products can integrate without guessing, duplicating logic or losing responsibility for their domain.
Speed is not the only latency
A query can run in milliseconds while a decision waits a week. Measure the full path from credible evidence to appropriate action.
Useful intelligence remains contestable
If people cannot challenge an analysis or inspect its assumptions, intelligence becomes authority without accountability.
Future-proof is a misleading promise
No system is immune to change. A better objective is to make change understandable, affordable and safe.