Loading
Practical notes from real engagements, including the mistakes and the parts that surprised us.
Most build-versus-buy decisions compare a licence quote against a development quote. That comparison is almost always wrong, because it leaves out the four costs that dominate years two through five.
Rakesh Thakur · 14 September 2026
The most valuable output of an AI feasibility study is often the recommendation not to proceed. Here is what we look for, and why finding it early saves considerably more than the study costs.
Vikram Sethi · 7 Sept 2026
Retries are not an edge case, they are the normal operating condition of any distributed system. Here is the design that keeps a duplicate request from becoming a duplicate charge.
Arjun Nair · 31 Aug 2026
Staging that does not match production is worse than no staging, because it produces confidence rather than information. Four differences account for most of the surprises.
Ananya Deshpande · 24 Aug 2026
Across assessments over the past two years, the same handful of issues account for most of what we report. None of them are exotic.
Ishita Roy · 17 Aug 2026
The difference between an assistant customers trust and one you have to withdraw comes down to three design decisions, none of which are about the model.
Vikram Sethi · 10 Aug 2026
Big-bang go-live is a decision to concentrate all your risk into a single weekend. There is almost always a better sequence available.
Manish Agarwal · 3 Aug 2026
Page speed erodes one release at a time. A number enforced in CI is the only thing we have found that reliably stops it.
Neha Bansal · 27 Jul 2026
A note on the name, and on the standard we are trying to hold ourselves to when nobody is checking.
Rakesh Thakur · 20 Jul 2026