Experiment / 01
A brief-diagnosis agent that maps a project's hidden constraints.
Used at the start of every engagement. Cuts our scoping time in half and surfaces the questions we should be asking the client on day one.
/lab
The room where we run our own systems.
The Lab is not a "play area". It is the operating environment we live in — the design tokens, AI workflows, and internal tools that run our studio. Every system we ship has been field-tested here first.
A studio that doesn't run on its own systems is a studio that has never tested them.
Most studio "labs" are showcases. Ours is the opposite: the things in the Lab are the things we depend on every day. They are visible because they have to keep working, not because they were built to be seen.
.02 / Domains
Four working systems in operation.
-
L.01
Internal design system.
The token library, components and editorial rules used across this site, our client surfaces, and our internal tools. Versioned, documented, and updated weekly.
-
L.02
AI co-operator workflows.
Prompt libraries and agents wired into the work we do every day — research summarization, brief diagnosis, editorial review. Each has guardrails and an evaluation harness, the same as those we ship to clients.
-
L.03
Internal tools.
Small operational interfaces we built for our own workflows: project diagnosis, deliverable review, feedback consolidation. Some of them have made it into client engagements.
-
L.04
Operating frameworks.
The frameworks we use to structure work — brief diagnosis, scope mapping, deliverable QA. Open enough that you can read them in the Knowledge section and apply them yourself.
.03 / In operation
A few of the experiments currently running.
These are not finished products. They are systems we run, observe, and revise every week.
Experiment / 02
An editorial QA system that reviews deliverables against our own standards.
The same system we ship to publishing teams. It first graded our own homepage — and rewrote two sections.
Experiment / 03
A retrieval system over our own knowledge base.
Indexed against operating questions, not topics. Answers reference internal frameworks the same way a senior would.
Experiment / 04
A handoff prompt-builder for AI work given to other tools.
Standardizes how we brief Claude, ChatGPT, or a designer. Reduces revision cycles by giving every operator the same constraints.
If a system works in the Lab for six months, it earns the right to ship. If it doesn't, no client will fix it for us.
.06 / Invitation