we.ather: Ambient social visibility for AI work
Lately, workdays have a strange shape. Many of us spend hours in deep, engaging collaboration — asking questions, building, testing — and at the end of it we might still not have talked to another human. The collaborator is an AI coding agent. The work is real, the interactions might feel uncannily real, but the room is empty.
This isn’t necessarily a complaint about AI tools. This rapid-pace augmented work feels exciting and ambitious. The risk I see is a side effect that wasn’t designed on purpose: the worlds we’re spending more and more of our time inside (the terminal, the agent chat) are the most private, sealed-off environments in computing. This is by design: when you’re sharing so much of yourself with a tool, privacy is paramount. But when your collaborator lives in the machine, the ambient social fabric and hum of work and making disappears.
The hum matters to our human, social-mammal brains. We know this from decades of research and from ordinary life: people work and exist near each other on purpose. Cafés, studios, labs, coworking spaces — even when nobody talks, presence does work. It regulates motivation (body-doubling is a whole practice built on this), it creates the openings through which curiosity and serendipity arrive, and it answers a question that turns out to be important for wellbeing: is anyone else out there doing this too?
The privacy-presence paradox
Of course, the answer isn’t just “make your work public.”
In my own research on social visibility in learning and work — exploring design probes like InquiryBits, which I’m presenting at CSCW this year — I found what I’d describe as a paradox held by almost everyone at once. People expressed a strong appetite for more visibility into each other’s work: they wanted the awareness and the openings for connection it creates. And in the same breath, they had very strong feelings about limiting that visibility to small, trusted groups. More visibility within a specific group, but not broadcast or surveilled.
The broader CSCW literature has circled this for thirty years — from the media-space experiments of the early 90s that piped ambient video between offices, to Erickson and Kellogg’s “social translucence”: the idea that systems should make social information visible but lightweight, enough for awareness and accountability without exposure. Most paradigms that have become industry-standard actually went the other way. We got the feed and the status dot. GitHub contribution graphs perform milestone-based productivity to strangers (and are increasingly AI-generated anyway). Slack presence dots tell your manager you’re “active” in a way that inspires mouse-jigglers and carries little meaning.
The tools for trusted mutual ambient presence — the café, but for the work we actually do now — are harder to find. How do we design for connection inside the expectation of privacy, not against it?
we.ather
For the past few weeks in a residency at Stochastic Labs, I’ve been experimenting with building an approach.
It’s a little tool called we.ather, and the metaphor describes what the feeling I want to capture: rather than momentary status, your work has weather, a shape and feeling over time built from individual moments.
A small sticky window shows a slowly drifting field of pixel clouds — one strip per “room,” a room being a handful of people (approximately 2 to 8 people) who have chosen to share presence with each other. My clouds and my friends’ clouds intermix as colored weather: color carries the kind of work (exploring, shaping, building, debugging, polishing, writing, etc), position carries time and topic-neighborhood, and that’s roughly it. When a friend starts working, clouds appear. You can have different rooms with different groups of people, and either see them individually or fused (in your view) into a denser sky.
This is deliberately not a dashboard or a feed. There are no names or metrics on the field, no history past the 4-hour weather window. If the room members allow it, you can hover a fresh formation to see a <5-word gist of what someone’s up to, and who it is (“caitlin · tidying utility code”) — this more detailed view fades shortly after it first appears, so clouds from an hour or more ago are ambient-only rather than hoverable. It should feel like glancing up for a moment at the room around you, rather than consuming a feed. If you stop working — or pause sharing, which is every member’s standing right — your weather simply ages out. Absence and privacy are indistinguishable by construction.
One of my favorite outcomes from this in the past couple weeks has been the increase in messages between my friends who are running their own we.ather boards. A tiny signal of visibility can spark curiosity and conversation that would otherwise have had no foothold, and enable those human-connection moments: “hey, what are you building?”
The privacy architecture
Because the core point is presence within the expectation of privacy, the architecture is intentionally minimal-risk. The short version:
Everything rich stays on your machine, in existing log architecture that this lightly taps into. The system reads the recent logs of AI-work sessions you explicitly allowlist — folder by folder, nothing parsed that you don’t opt-in on — and a small model distills them locally into low-resolution state: a work-phase and a few topic words. Your actual chats with your tools never leave your laptop.
Only the weather is shared. What crosses the network is roughly one short line of state e.g. “building · improving setup documentation”. Each room has its own resolution dial: a room can share topics, or presence only (color-coded clouds, but no hoverable topic text); all room members can see what the settings are.
Rooms are self-hosted, not platformed. There is no server of mine in the middle. Each room’s tiny relay runs on the room host’s own free Cloudflare account; it holds only the low-res states, for seven days, and its operator is a disclosed member of that room. My friends’ rooms with their friends run entirely on their infrastructure: I can’t see them, and there’s no aggregate anywhere of anything.
Leaving is quiet. Pausing purges you from your rooms within a cycle, and to everyone else it just looks like still air.
Want to try it?
I’ve been piloting this with a few friends, working in different AI stacks. This pilot is tiny on purpose, but it’s ready for a few more people.
[Update — you don’t need to message me anymore, you can install this directly from the latest release on Github! I’d love to know if you’re using it and how you feel about it.]
If you spend time working with AI tools (Claude Code, Codex, or Warp right now; more later, or let me know what you’re using!) on a Mac, and you have 1–8 people you’d actually want mutual ambient presence with — send me a message. You can set up and own your own we.ather board (I don’t need to be in the loop). Setup is ~10 minutes, costs pennies of model API per workday, and I’m happy to share the privacy architecture docs for transparency and feedback.
I’d love to make we.ather with you.
What a brilliant and fun idea. Kudos!
First, this is fun and fascinating. Next, presence has always had a certain Icarus quality, too easy to fly too close to the sun.
We live in a time when the most interesting ideas reliably become the most disruptive.
Presence is helpful. It’s what we miss. It’s also a problem.