HOW WE WORK
START WITH FRICTION.
MAKE SOMETHING USEFUL.
AI is a building material, not the point.
We care about what helps people do the work. Sometimes that means AI. Sometimes it means conventional programming, a better workflow, a form, a spreadsheet, or an integration. Use an appropriate amount of technology. The spreadsheet does not always need a successor.
From “this is annoying” to “this helps.”
Notice friction.
Start with a recent example: the repeated spreadsheet, the unclear request, the handoff that sends everyone back to their inbox. What actually happened matters more than the name of the solution.
Understand the problem.
Follow the work from one person or tool to the next. Identify missing context, repeated steps, and decisions that still need a human. Agree on what a useful result looks like, including access and information boundaries.
Prototype the useful part.
Choose a small piece worth trying. It might be a form, spreadsheet, integration, conventional code, or an AI-assisted draft. Use a sample to make the idea concrete before building around assumptions.
Test what matters.
Try the actual task. Check the sources, unclear cases, keyboard controls, and small screens. A generated answer is a draft to examine, not evidence that the job is done.
Refine with feedback.
Watch where someone gets stuck. Change the wording, the flow, or the underlying approach. Quick prototypes make it easier to learn; they don’t remove the need for judgment. Sometimes the useful change is three steps earlier.
Deliver—or take another pass.
Review the working result together. Make the instructions, ownership, limits, and unresolved items clear. If it still doesn’t solve the problem, keep iterating or stop building the wrong thing.
THE METHOD IN THE WORK
Make the result easy to question.
The existing tools show what that means in practice: source-linked findings, unanswered questions, editable drafts, and a review before sharing. Speed helps us reach something testable sooner. It doesn’t make uncertainty disappear.
Public Workbench experiments let people try an idea. Practical solutions can instead be built around one team’s work, access, and responsibilities. Related thinking; different jobs.
Start with your problem ↗