A small lab for trying things with AI. We build quick experiments, see what holds up, and turn the good ones into tools people use.
Run your cursor across the nameDrag a finger across the name
How an experiment runs
-
Step 01
Tinker
Start with one question and a rough prototype. Most take days, not months.
-
Step 02
Test
Put it in front of real people and real data. Write down what happened, including the strange parts.
-
Step 03
Shift
Decide what it becomes: a tool worth keeping, a new direction, or a lesson for the next one.
Every experiment gets a status
- Running
- Still on the bench. We poke at it every day and change it often.
- Shipped
- Held up under real use and grew into something people rely on.
- Shelved
- Didn't work out. We keep the notes, because the next idea usually starts here.
What we like to build
Single-purpose models
Language models set up to do one job very well.
Workflow agents
Assistants that take over the repetitive steps in real work.
Making tools
Playful tools for working with images, sound and words.
Tiny useful apps
Small apps that answer one question fast and get out of the way.
Have something worth testing?
Send a problem, an idea or a collaboration pitch. Every note gets read, and we reply by email.