Notes on infrastructure, operations and AI.
Short pieces on infrastructure, observability, cloud foundations and how AI can support better operational decisions in complex technical environments. Written from the perspective of someone who has had to keep these systems running, not only design them.
The code is cheap. The reading is not.
AI writes code that compiles and reads as competent. None of that is a security property. The writing got cheap, the reading did not, and the reading is the control.
The price is not the cost
An AI business case is priced on a number you do not control: the cost per useful outcome. The price per token falls while the work per task rises, and today's flat rate is being held below cost.
Failover assumes a copy
Every failover plan assumes the backup is equivalent to the primary. Networks make that true by design. A backup model cannot be a copy, which changes what redundancy actually buys you.
Fluent is not the same as correct
A model is built to produce writing that reads as though a competent person wrote it. Whether the claim matches reality is a separate question, and checking it is the job now.
The model inherits the foundation
A model does not rise above the quality of what feeds it. It absorbs that quality and passes it on. In operationally serious environments, that makes the foundation the whole game.
Build observability for the worst moment
A wall of green tells you nothing the minute it matters. The only real test of observability is whether a person can use it under pressure, when there is the least time and the most at stake.