Name the idea sharply
Every build begins with a single, narrow question. What is the friction? Who feels it? What changes if it disappears? Idea first, surface second, code last.
Step 1 of 3An atlas of the possible.
Spring 2026 — six ideas, borne into action through AI. Some are studies, some are shipped products. Ideas finally taking shape.
Plate I — A First SketchEvery build begins as one sentence written on paper. If the idea can’t fit on a single line, it isn’t ready yet — and we keep editing the line until it is.
Then we ask the harder question: what changes if it works? For whom, in which workflow, on which Tuesday afternoon. That answer becomes the brief — concrete enough to design against and small enough to ship.
From there the work is unromantic. Build the smallest thing that proves the brief. Put it in front of the person who needs it. Watch what they actually do with it. Iterate.
A working method for turning a sharp idea into a running system. Three movements; no more, no fewer.
Every build begins with a single, narrow question. What is the friction? Who feels it? What changes if it disappears? Idea first, surface second, code last.
Step 1 of 3Most ideas don't need an app — they need a quiet agent, a small script, an automation. We cut to the smallest thing that proves the thought, then we ship it.
Step 2 of 3AI is a material. We compose models, tools, and code into loops that learn and improve — and we keep the seams visible so they stay honest, debuggable, and yours.
Step 3 of 3“The business of every art
is to bring something
into existence.”
Two builds from the spring run — one a study still finding its shape, one shipped and quietly running.
An early sketch. A planning tool for a major Canadian university, turning the course catalog, schedule, and campus map into a single workflow built for first-year students. Deliberately narrow: one program, one entry point, one set of constraints to honor.
Deployed. A native desktop tool that captures both sides of a real conversation — locally, on the operator's machine — and transcribes them with whichever AI model they pick that day, cloud or fully on-device. No bot in the call, no vendor pipeline, no per-seat tax. Audio and keys never leave the laptop.
Apps, agents, and automations — built for people, teams, and organizations. We take a small number of engagements each season.