Half of what we make
is not on any roadmap.
Innovation, for us, is not a slogan on a wall. It is a practice — a way of paying attention. Some of it becomes a product. Some of it becomes a paper. Most of it becomes discipline for the work that will come after.
Four practices.
We do not have an innovation team. We have four practices we all return to, whatever we are building.
Read the signal, not the news.
We follow research, not press releases. We read papers, watch prototypes, and pay attention to the behaviors that are quietly forming underneath the surface of a category.
Build the smallest version.
Every hypothesis becomes a running prototype within a week. Not a slide. Not a mock. A thing you can open, use, and be surprised by.
Talk to five real people.
Nothing replaces sitting next to a real user and watching them try the prototype. Five conversations are worth more than five surveys.
Keep a slow notebook.
Not every insight is ready to be a product. We keep a long, slow notebook of things we noticed — and we come back to it. Half of what we ship first appeared as a line in a notebook years ago.
Three time horizons, at once.
At any given moment, some of our attention is on now. Some on the next twelve months. Some on the years past the horizon. All three are treated seriously.
Ship the next thing.
What is already in front of a real user, receiving real love — and how do we make it kinder, calmer, and more capable this quarter. The largest share of our energy always sits here.
Prototype the next chapter.
What could be true a year from now if a few things break our way. Small teams, small budgets, small honest bets. Some of it will graduate to product. Some will not.
Sit with the next century.
What we hope will be true in a decade. What kinds of companies, tools, interactions, and products the digital century needs. This is where we read, write, and think out loud with people we admire.
Four quiet experiments.
A short, honest window into what we are working on that hasn't shipped yet. Some of it will. Some of it won't. All of it teaches us something.
Presence that listens.
A study of what it feels like when software knows you are there without asking — attention as a channel, not a metric. Early. Uncertain. Interesting.
state · explorationSlow intelligence.
Models that think for minutes instead of milliseconds — and interfaces that make the wait feel like care, not friction. Aimed at decisions that deserve weight.
state · prototypeConsent by default.
An identity layer where the user is the only party who can grant, revoke, or renegotiate what a product knows about them. Boring on purpose. Load-bearing by design.
state · draftRooms that compose.
Real-time audio and video primitives that treat conversation as a first-class object — recordable, replayable, transformable. A quieter, more human take on live software.
state · closed researchWorking on the same frontier?
We collaborate with a small circle of researchers, founders, and labs. If you are on nearby ground, come talk to us.
Share your work →