AI decisions
When AI is actually useful, and when it isn't
A working test for whether a problem wants a model, a rule, a query, or a conversation with the person who owns the process.
Insights
Reasoning about technology decisions.
Short pieces on the decisions that come up most. They contain reasoning, not statistics we can't stand behind.
Technology economics
Build vs buy is four decisions, not one
Four costs, and why the option that wins on build usually loses on at least one of the others.
Architecture complexity
Do you need Kubernetes? A complexity-budget answer
Kubernetes is an excellent answer to a specific set of problems. The question is whether you have those problems.
AI operating cost
The AI cost that arrives every month
Inference, retrieval, evaluation, re-runs and the human review loop: how to model the operating cost of an AI feature before you build it.
AI agent architecture
AI agents: a case for restraint
Most “agent” problems are workflows with one or two model calls in them. Knowing the difference is the architecture decision.
Decision traceability
Architecture decision records that actually get read
An ADR is only useful if the next team can find the reason. Eight fields, one page, and a rule about assumptions.
Technology reversibility
Reversibility: the dimension most decisions skip
Two decisions can have identical cost and value and still deserve completely different amounts of deliberation.
Topics we write about
- When AI is actually useful
- When AI is unnecessary
- Build vs Buy
- Architecture complexity
- Kubernetes decisions
- Microservices decisions
- AI agent architecture
- AI operating cost
- Cloud economics
- NFRs
- Architecture decision records
- Legacy modernisation
- Technology debt
- Vendor dependency
- Technology reversibility
Have a decision you'd like reasoned through? Bring us the problem
Free · 30 minutes · one real problem
Bring a problem. Leave with clarity.
Thirty minutes, one real problem, structured thinking. If there's no value, there's no engagement.