The rewrite hiding under the roadmap
How to spot, before the committee meets, that a product plan quietly assumes an architecture the company has not built yet.
Everything here comes out of real engagements, with the identifying detail removed. If a finding turns up in enough estates to be a pattern rather than a coincidence, it ends up on this page.
Across our last set of inventories, the number of AI systems actually running in production consistently exceeded the number on the client register. The gap was not shadow IT in the old sense. It was teams doing exactly what leadership asked for, quickly, with the credentials they had to hand.
How to spot, before the committee meets, that a product plan quietly assumes an architecture the company has not built yet.
Power, cooling, network, licences, spares and the people who run it at three in the morning. Where the three year number actually goes.
Why so many production evaluations overlap their own training set, and how to build a held-out set that actually tells you something.
The most common argument we have with clients, and the reason we will not resolve it in their favour.
Rack density and cooling method constrain everything above them, and they are the layer most programmes discover last.
Reading commit history and interview notes together to find the key person risk that a headcount plan will not show you.