Insights
The positions
behind the writing
Editorial note
We publish a small number of articles and we would rather they argue something than cover a topic. This page is the lens on them: a handful of themes, each with the position we hold, and the articles underneath that make the case.
Nothing here is new writing. Every article listed lives on the blog, which is the canonical, chronological archive and the place to go if you want to read in date order or filter by category.
8 articles published across 4 themes.
01
Grounding decides what a system can say
3 articles
Most wrong answers from an AI system are not model failures. They are retrieval failures, scoping failures, or a build that was never told what it is not allowed to answer. We write about this more than anything else because it is where the difference between a demo and a usable system actually sits.
02
Design for the day it fails
3 articles
An automation that works on the happy path is half a system. The other half is what happens on a timeout, a bad payload, an ambiguous case or an answer the model should have refused. Escalation paths, instrumentation and alerting are architecture, not a phase that gets cut when the timeline tightens.
03
One source of truth, or none
1 article
The fastest way to damage an operation is to let a new layer start keeping its own copy of the customer, the order or the record. Before anything is integrated, the system of record has to be named and the write path has to be deliberate.
04
Other field notes
1 article
Writing that does not sit under one of the themes above. Listed here rather than dropped, because the themes describe what recurs, not everything we have published.
The full archive
Everything we have published, newest first, filterable by category. If you came here looking for a specific article rather than an argument, start there.
Want one of these applied to your own system?
Book a strategy call. We will tell you which of these problems you actually have and which ones you do not.