Leisure Labs is an AI research and product lab. We build systems that reduce unnecessary work and expand what people are able to do.
For most of history, only a few people had the time to learn, to study, to make things carefully, or to take part in public life. Everyone else worked. The promise of labor-saving machines was never only efficiency — it was the wide distribution of time, the basic material of human development.
That promise is still unfinished. Much of modern work is friction: coordinating, translating, retrieving, reconciling, repeating. And much of modern automation makes people more passive rather than more capable — faster at producing, worse at understanding what they produce.
We think intelligent systems should be held to an older, higher standard: not only what the machine can do, but what people are newly free to do. We design for returned time, build tools that deposit skill rather than absorb it, and study the connective infrastructure that removes friction between people altogether.
We call this leisure in its oldest sense — not idleness, but unforced time used well: for learning, care, craft, reasoning, and participation in common life. It is what machines are for.
The first promise of any labor-saving machine is hours. Kept honestly, that promise is the most valuable thing technology has ever delivered: the eight-hour day, the weekend, and the schooling years of childhood were all paid for, in part, by machines doing work that people once did.
We hold AI to that original standard. A system that completes tasks while its users end the week with no more time — for judgment, for learning, for one another — has optimized the wrong variable. We instrument for returned hours the way others instrument for throughput.
There are two kinds of tools. One kind absorbs skill: use it for a year and you can do less without it than when you began. The other kind deposits skill: the spreadsheet taught a generation more applied mathematics than any curriculum. The difference is not the task. It is the design.
We build and evaluate for the second kind. Our systems expose their reasoning, route judgment to the person and repetition to the machine, and count it a failure when capability quietly migrates out of the human and into the product.
Understanding compounds when it is shared and stalls when it is hoarded. We publish our research, release our evaluation instruments, and document our infrastructure, because a finding that cannot be examined, reproduced, and built upon is not yet knowledge.
Where safety or privacy requires restraint, we say what we are withholding and why. Quiet is acceptable. Opacity is not.
Almost everything difficult that people accomplish, they accomplish together — yet most AI arrives shaped for a person alone, and a tool that speeds up every individual can still make the group around them dumber. We design for the room: shared context the whole group can see, provenance on every claim, disagreement surfaced rather than averaged away.
The test of a system is whether the group becomes smarter — better at remembering, arguing, and deciding — not merely faster at producing.
The technologies that have saved the most human effort are mostly agreements: standard measures, shipping containers, common protocols, shared language. They work by removing friction between people — permanently, and for everyone at once.
We treat the connective layer of the AI era — how systems cite, exchange, version, and trust knowledge — as a first-class research object, because that quiet layer will decide how much of everyone's time the next decade actually returns.
We are a research lab that ships. Our tools go into real working rooms — study groups, small teams, workshops — and what happens there flows back into the research as evidence. We measure people, not just models.
Building keeps the thinking honest. A principle that cannot survive contact with a Tuesday afternoon's actual work was not a principle. It was a press release.
AI is evaluated by whether it completes tasks. We argue for a complementary standard: whether it expands the time and capacity people have for judgment, learning, care, and shared work.
Amara Osei, Daniel Reyes, June Park
Most AI interfaces assume one person and one model. Most meaningful work happens in groups. Design patterns from a year of building shared-context systems for small teams.
June Park, Tomás Carvalho
A benchmark tells you what the model can do. It tells you nothing about what its users can still do. First results from paired capability–capacity evaluations.
Daniel Reyes, Priya Raghavan
A lab built on one premise: the point of intelligent machines is not what they can do, but what people become free to do.
Not all work is worth saving from automation, and not all of it is worth automating. The category that matters is the work that should not exist.
The technologies that save the most human effort are rarely tools. They are the agreements that let everything else connect.
We are a small lab looking for people who care about both technical depth and the human consequences of tools.