Engineering
Legacy modernization without the big-bang cutover
The riskiest way to replace a critical system is all at once, on a date chosen a year in advance. There is a route that looks slower on a plan and finishes sooner.
Insights
What we have learned building this stuff
Notes on modernization, applied AI, cloud and the operational work that decides whether any of it lasts.
Artificial Intelligence
What an AI agent actually needs before it reaches production
The prompt is the easy part. Permissions, reversibility and a real evaluation set decide whether an agent is safe to let near a live system.
Engineering
Five questions worth answering before a cloud migration
Most disappointing migrations were decided at the wrong altitude, as one programme rather than a series of per-workload decisions.
Engineering
Why your integration keeps breaking at month-end
Integrations are usually tested against a system that is healthy and idle. They fail against one that is slow and busy, which is what month-end produces.
Artificial Intelligence
When not to build a model
Half the value of an AI assessment is the list of things it tells you not to build. Here is what usually ends up on it.