Working principle
AI Should Reason, Not Own Authority
How to use models for investigation, hypotheses and planning without handing production authority, policy and approval to a probabilistic system.
Ideas
Ideas tested in real systems before they become essays.
Writing is a way to make judgment visible: what is authoritative, where failure lives, what should be simplified, and where AI should remain an advisor.
- AI × Authority
- Platform × Product
- Failure × Recovery
- People × Ownership
- Runtime × Truth
Featured thesis
AI Should Reason, Not Own Authority
AI can synthesize evidence, form hypotheses and propose plans; policy, approval and execution should remain outside the model, deterministic and auditable.
See the thesis in a real systemIdeas I work with
Not topics. Decision lenses.
Working principle
Control Planes That Tell the Truth
Why control planes must separate intent from observed runtime, and why optimistic status can be more dangerous than explicit failure.
Working principle
Build Systems. Build People.
Technical leadership scales when architecture, context, ownership and new technical leaders are built together.
Working principle
Technology Strategy Is Organization Strategy
A technology roadmap should not start with trends and tools; it starts with organizational direction, capability gaps, constraints and build, integrate or defer decisions.
Working principle
Infrastructure Becomes Valuable When It Becomes a Product
Infrastructure creates leverage when authority, APIs, workflows, lifecycle, governance, recovery and user experience are designed together.
Working principle
Design for Failure, Not the Happy Path
Why topology, durable operations, fencing, resume and recovery belong in the design from the start rather than being appended after incidents.
Working principle
Why Operational Truth Must Come From Runtime
The difference between source implementation, validation, installed runtime and observed state — and what each can legitimately prove.
Working principle
Simplification Is an Architecture Decision
Why removing duplicate stacks, extra authorities and unused abstractions often improves reliability and comprehensibility more than adding technology.
From field note to principle
An idea matters when it improves behavior.
A useful framework should help a team decide better under ambiguity, not just produce a good sentence for a slide.
- ObserveObserve real behavior and constraints.
- Name the boundaryName authority, state and ownership.
- Make the trade-off explicitMake the cost of choosing and not choosing explicit.
- Test against runtimeThe idea has to survive contact with runtime reality.