Perspective
Technology
What makes technology useful?
A stricter test than a feature list.
Product
The software after software.
When software can act, the product is the outcome.
Building
Why AI products need context.
Models are cheap. Memory is not.
Artha
What does it mean to build something that matters?
A refusal, not a slogan.
Practice
Taste is a systems problem.
Headcount is not a strategy for judgment.
Craft
The artefact is the argument.
If it cannot be used, it has not been said.
Common questions
Timelines, ownership, data, and what we will not take on. If yours is not here, write. Two working days. Usually sooner.
Everything worth asking first.
Write to usOne team, one conversation. Strategy, design, engineering and AI in a single pass rather than a relay between vendors. The people who make the decisions are the people who ship the work, and they stay reachable after it ships.
A first working artefact in weeks, not quarters. We would rather put a narrow thing in front of real users early than present a wide thing that has never been used. The full engagement length depends on the problem — we will tell you our honest estimate before you commit.
You do. Code, models, infrastructure and documentation are yours outright at the end of the engagement. We do not retain licences on client work, and we do not build on a platform you cannot leave.
No. “Add AI” is a prompt, not an engagement. If the underlying process is unclear or the data is not there yet, the honest first piece of work is naming that — and we will say so rather than build a demo that impresses and changes nothing.
It stays yours and it stays where you put it. We build against your infrastructure and your permissions model, and we prefer architectures where sensitive context never leaves your boundary. Any exception is a decision you make explicitly, not a default we choose for you.
Yes, and that is usually the better outcome — the goal is a way of working the organisation keeps once we are gone. What we will not do is staff a bench. We are not extra hands on someone else's plan.
One test: would the absence of this thing be expensive? Not whether it is technically impressive, not whether the category is loud. That question is how we say no, and it is the reason the product list is short.