Short, practical notes on deciding where AI belongs, how to make it dependable, and when a simpler system is the better answer.
A reference design for capturing agent state through MCP or an API, then restoring it into a controlled runtime after failure, migration, or operator intervention.
9 min read
Read articleA practical guide to the trade-offs between high-precision weights, Q8, Q4_K_M, IQ2_XXS, and ternary models on consumer hardware.
6 min read
Read articleA security-focused approach to testing what an agent can actually do, with real isolation, default-deny networking, safe fixtures, supervised runs, and evidence that shows when a boundary holds.
12 min read
Read articleAgentic systems need bounded responsibilities, useful tools, and a clear answer for what happens when the model is wrong.
4 min read
Read articleRestructuring around AI is an operating decision: it changes processes, team responsibilities, tooling, and culture together.
5 min read
Read articleEvaluation gives a team the evidence needed to measure, monitor, and improve an AI system in production.
4 min read
Read articleAI rationalisation is a practical review of cost, reliability, compliance risk, and whether a language model is the right tool at all.
4 min read
Read articleGiving an agent a way to pay is easy. Defining its authority, smart-contract controls, and what happens after settlement is the real systems work.
6 min read
Read article