Thinkery · Ideas about AI
Thinkings
How we think about the AI we build.
Seven essays on the choices behind our products: what efficiency means, how AI uses knowledge and why control over data matters.
Our starting point
What an AI answer really costs
Time, computing and the work of correcting an answer all count. Why we measure efficiency by the completed task.
Read the essayExplore the other essays.
02 — 07AI needs the right context
Which documents, rules and decisions does AI need for a task? That question comes before the first prompt.
Sep 2026What AI should remember
Useful memory keeps decisions with their reasons, and allows them to change as we learn.
Sep 2026How a question becomes a product
Which function do we need? How do we test its value? A look at the method behind our product decisions.
Sep 2026Your data. Your choice.
Where is information processed, who may use it and what can you take with you? Making data sovereignty practical.
Sep 2026When AI works with AI
Shared work needs clear assignments and reliable handovers. AI systems are no exception.
Sep 2026Clarity belongs in the product
A click should have a clear meaning. Why language and design help determine whether software does its job.
Sep 2026