INSIGHTS
Notes from the build.
What we learned shipping AI, data and platform systems into production — including the parts that did not work. No thought leadership, no predictions, no hype.
Browse by topic
- AI6 minYou cannot ship AI without an evaluation setEvery stalled AI project we have seen stalled at the same question: how do you know it works? Here is the cheapest useful answer.18 Aug 2026
- Data5 minThe recommendation bug you cannot seeOffline metrics look fine. Online behaviour does not match. The usual culprit is that your model sees a different world in production than it learned from.29 Jul 2026
- Engineering6 minReach for a queue before you reach for clevernessMost throughput problems are solved by separating work into durable units, not by optimising the work itself.10 Jul 2026
- Engineering5 minWhat answer engines actually need from your siteA growing share of queries are answered without a click. Being the source that gets cited is a structural problem, not a content-volume one.24 Jun 2026
- Practice4 minWhen not to use AIThe fastest way to lose credibility with an AI project is to apply it where a deterministic system would have been better, cheaper and correct.5 Jun 2026
6 posts shown for All.
Got a harder version of one of these?
These notes come out of real engagements. If one of them describes your problem, we should talk.