Social infrastructure for the age of AI.
AI is making coordination cheap. Markets will use that to get better at extraction. We want to use it to help people form circles. Machines will probably out-think us; what matters more is whether we still know how to be together. The loop we're building: AI notices, circles gather, spaceholders hold them, service follows, stories travel, and the whole thing deepens.
This roadmap covers one strand of the ServiceSpace ecosystem, the AI-assisted weaving. Most of what ServiceSpace does has never had a roadmap and doesn't need one.
Every new technology changes what people can do together. AI is changing it again, and that leaves us with a choice.
For most of modern history, getting people to act together has been expensive. It took institutions, managers, meetings, markets, platforms, and a lot of overhead. Ronald Coase made this point in 1937: firms exist because coordination has a cost. Each major technology has shifted that cost. Railroads and the telegraph made continent-wide companies possible. The internet made distributed markets possible. Platforms then optimized for audiences, transactions, gigs, and clicks.
AI shifts it again. Software can now plan, summarize, translate, match, schedule, remember, and follow up. None of that is entirely new. What's new is that the help is always on. One person can do what used to take a team, and a loose network can respond almost like an organization.
If AI just speeds up the way markets already work, we'll get smoother transactions, sharper persuasion, and more efficient extraction of attention. People will be more connected to hubs, brands, and platforms, and less connected to each other.
We already know this pattern. Most organizations know how to gather an audience. Few know how to turn one into a community. The default is one-to-many: one speaker, one platform, one newsletter. People are linked to the center but not to each other, so when the center goes quiet, so does everyone else.
That design is efficient but fragile. It lowers transaction costs and raises relational ones. You get reach without belonging, and a lot of goodwill goes to waste because there's no path from being inspired to knowing someone to doing something together.
This usually isn't bad intent. Plenty of builders sincerely want community, but anything funded by extraction eventually has to feed the hub. The odd thing is that AI is now making many-to-many coordination nearly free, so in principle nobody has to be stuck in that pattern. It still hasn't changed, because the skills of building real community have gone out of practice. What's missing is someone outside market pressure, focused on exactly those skills, who also knows how to use AI well. That's an unusual combination. It's also a buildable one.
So the real opportunity isn't automating more tasks. It's lowering the cost of coordination that serves relationships, and then stepping back. AI can match people on what they care about, not just demographics. It can read thousands of reflections and notice shared questions no facilitator could track alone. But if it stays in the middle, it becomes the new hub.
Any large platform could say that sentence. The difference is which way things flow. In a market model, relationships end up feeding the platform, and every human interaction is judged by what it returns. Here the software serves the relationships and then gets out of the way. The real test is who's still in the room at the end.
AI will keep getting better at what can be measured. Collective Heart Intelligence (CHI) is about what can't be.
Over the next few years AI will keep extending into prediction, analysis, persuasion, and productivity. Whatever we call the most capable systems, machines will do more and more of what we used to call thinking.
But some intelligence doesn't live inside any one mind. It shows up between people: when a room starts really listening, when strangers become a circle, when generosity moves without anyone managing it, when there's enough trust that disagreement leads somewhere useful.
We call this Collective Heart Intelligence, or CHI.
AI runs on big data: huge records of what has already happened. CHI runs on what we call deep data: context, conscience, care, timing, and trust between particular people at a particular moment.
Why does this matter so much? Because even the most hopeful visions of AI depend on it. AI might help people live longer, give us more free time, and support creativity and ecology. Those benefits only help if people are connected to each other. Free time in a lonely world turns into distraction. Abundance without trust turns into hoarding. CHI isn't a competitor to AI. It's what lets AI's benefits actually reach people.
So the bet of this roadmap is this: as AI speeds up the measurable world, we need to build social infrastructure for the parts of life that can't be measured.
By infrastructure we mean real things: circles people can join, service people can offer, stories people can pass on, and a way of practising together that others can copy without anyone owning it.
This roadmap grows out of twenty-seven years of ServiceSpace practice, with no fundraising campaigns and no ads.
That history matters because you can't build this from scratch. It depends on trust. ServiceSpace has spent decades learning how to hold spaces where people move from inspiration to reflection, from reflection to connection, and from connection to service, through newsletters, Awakin Circles, KarmaTube, DailyGood, pods, Laddership, retreats, and a great deal of volunteer work nobody sees.
It has also always been a technology story. The pod platform already handles registration, events, peer reflection, translation, video calls, email, volunteer coordination, and community memory. The difference is what it's for: not conversion or retention, but helping people find each other. The technology serves the relationships and then stays in the background.
Organizationally, ServiceSpace skipped a step, the way some countries went straight to mobile phones without building landlines. It never built a hierarchy it would now need to take apart. It has been a distributed, trust-based network from the start.
Subscribers, volunteers, circle hosts, retreat alumni, partners, and quiet friends, gathered mostly by not asking anything of anyone.
Thousands of experiments in what helps a room come alive, what deepens listening, what breaks trust, and what helps people move from "me" to "we."
The engineers have sat in the circles. The goal isn't to keep people on the platform. It's to help them show up more fully in their own lives.
It's unusual to have all three in one place, in a gift-based, non-commercial setting that has been tested in practice. That's the soil this grows from.
For years we could convene circles, host retreats, run pods, and support volunteers. But a lot of things stayed expensive:
So most of the world chose easier patterns: broadcast, platforms, funnels, campaigns, donor lists. Those scale quickly because they don't need to know people well. They only need to move them efficiently.
AI changes the cost. It can now handle the background work that kept many-to-many relationships from growing: noticing, remembering, matching, inviting, following up. That used to take too much central staff time. Now it doesn't.
The point isn't to replace human care with AI. It's to use AI to clear away the friction that keeps care from finding its next step.
The 5% is scaffolding: invitations, matching, reminders, ways in. The 95% is what nobody can manufacture: trust, sincerity, friendship, silence, courage, grace.
Most technology tries to own the 95%. It wants to keep users inside the system, sit in the middle of more relationships, and collect more data. We try to do the opposite: build just enough structure for people to take it from there.
Wherever someone wants to do something good, the beam's job is to notice it, support it, and connect it to others, quietly. That doesn't mean keeping AI in the back office. It means letting AI help at sensitive moments, like forming a circle or making an introduction, and then stepping back.
In practice:
This isn't anti-technology. It's technology with limits. Use software where it's strong: patterns, logistics, memory, translation, coordination. Stop before it takes over what only people can hold for each other.
Every part below is running or being built now.
"Stack" is a convenient word, but these aren't levels with circles at the top. Awakin Circles, DailyGood, KarmaTube, pods, and retreats are each complete in their own right, and people are changed by them without ever going further. What the newer agents add is movement between these offerings, so people can find each other across them. Some people will move around a lot; some will never touch an agent at all, and that's fine.
Contemplative companions grounded in wisdom traditions, teachers, and practices, answering tens of thousands of questions so far. Think of a well-read librarian more than an oracle: happy to help with the question, claiming none of the wisdom as its own. Sometimes an answer is enough. Sometimes what the person really needs is other people.
It starts one-to-one, and can lead to many-to-many.
On a typical platform, this data would become engagement analytics. Here it shows what people are wrestling with, where they overlap, and which circles might want to form.
Proposes circles of five to seven people, drawing from pod alumni, companion conversations, newsletter readers, retreat networks, or any group that opts in. Millions of people may be moved by a story or a call, but without a circle most of them stay an audience. With one, a shared moment can turn into a living room.
The agent proposes. People decide. Spaceholders hold.
Three calls, simple prompts, light support. The best sign of success is what happens after the structure ends. In the last pilot, most circles kept meeting on their own after the formal calls were over.
Do people keep meeting, serve together, start holding circles themselves? That's how we know it worked.
It helps organizers describe needs as small, personal opportunities, and matches people by what they can offer, when, and how ready they are. The aim isn't efficient volunteer management. When service is voluntary and comes from care, it tends to give people energy rather than drain them.
It comes second on purpose. Relationships aren't a means to service, and service isn't how we measure relationships. But service that grows out of friendship feels different from service assigned to strangers. Circle first, then the offering.
It doesn't create generosity. It gives it somewhere to go.
Being non-commercial matters in practice. A company built on extraction can't really build tools meant to fade into the background, and a company that needs a moat can't easily give its work away as a commons. Shunya exists to build tools whose job is to become invisible.
Shunya Labs is its own nonprofit, incubated outside of ServiceSpace, so the engineering can be funded and sustained independently.
If AI makes coordination cheap but people don't grow in depth, we just get shallowness faster. Laddership trains spaceholders, weavers, and volunteers in a different kind of leadership: not command-and-control and not charisma, but the quiet ability to help others rise.
It sets the pace for everything else. Each rate limiter in the roadmap below comes back to it. Spaceholders are a deliberate bottleneck. Plenty of this work stays human, but holding a circle is where we draw the clearest line: AI can assist, and must never substitute. That's why training spaceholders is the first and longest-running build in the roadmap.
Without Laddership, this is software. With it, it's social infrastructure.
Five steps, each leading into the next.
People show what matters to them all the time: in reflections, questions, companion conversations, pod posts, and moments of "I don't know why, but this moved me." Most of that used to disappear. AI can now notice that these seven people are asking similar questions, that this person keeps saying they want to help, that this story is ready to be shared.
More data would make the matching sharper, and connecting to inboxes and social networks is technically easy now. We don't. Circles are formed only from what people choose to share inside our programmes, with clear consent, not from whatever a crawler could collect. Good matching should come from listening well, not from gathering more.
We also try to remember that the data is never the whole picture. However detailed it gets, what happens in a pod between the words, or in a circle's silence, doesn't show up in it. The agent works from traces; the circle works from being there. The people building this need to keep that gap in mind.
A typical platform asks how to keep each user engaged with the platform. Circle Agent asks who this person might grow with. It forms circles in five ways:
Circles are ancient. What's new is being able to notice, match, invite, support, and remember them without a large staff. The bar to join stays low: consent and sincerity. Someone new can start by answering a few honest questions, or by bringing six friends.
A group chat isn't a circle, and a cohort isn't a friendship. Spaceholders look after the conditions: humility, deep listening, shared ownership, and not over-managing what emerges. AI can help a host prepare, suggest prompts, summarize context, and handle admin. What makes a circle work is still human presence.
Lots of people can build matching algorithms. Far fewer know which matches lead to trust rather than consumption, honesty rather than performance, depth rather than networking. The goal is noble friendship, not just connection.
When a circle deepens, people usually want to help. They say "How can I help?" or "I have a few hours" or "I don't know what I can offer, but I want to be useful." Today that energy often fades, not for lack of goodwill but because there's no obvious next step.
Someone might translate one paragraph, clip one video, welcome one newcomer, listen to one story, or host one circle. These small acts matter because they start relationships, not just because they get things done.
For example: a host mentions a backlog of untranslated Awakin passages. The agent remembers that someone in a circle reads Hindi and once offered her Tuesday evenings. It makes one introduction. That's the whole product. The rest is between the two of them.
In a market, a micro-task is a unit of productivity. In a gift economy, a small act is a way of caring for someone.
Stories need to travel as invitations, not advertisements. The goal isn't to prove impact; it's to make goodness easier to see and easier to join. One story can bring seven people together, and those seven can become a circle, a project, and eventually something others build on.
That's how CHI grows: by deepening a pattern, not by scaling a platform.
The arrows show a rhythm, not an assembly line. People may be in several circles at once, arrive one way and serve another, drift off, rest, and come back. Circles finish, dissolve, and re-form. What brought a circle together is where it starts, not a box it has to stay in.
Each year has a question, things to build, signs of success, and a rate limiter, because this grows at the speed of trust, not the speed of servers.
Make Metta Circles boringly reliable inside our own ecosystem. Test whether AI can spot shared interests, propose circles, and take on logistics while leaving the circles more alive, not less.
Open the pattern to people who already trust us: Gandhi 3.0 alumni, retreat friends, Awakin hosts, partner organizations, and networks already focused on service. Growth through trust, not marketing.
Let circles grow new ones: alumni becoming spaceholders, spaceholders becoming weavers, acts of service becoming stories, and stories bringing new people in. This is where it starts to spread on its own.
By 2029, many systems will use AI to optimize for speed, persuasion, and control. More arguments about what's wrong won't help much. A working example of something else might.
Two things shape this year. First, the tools of 2029 will be better than anything we can picture now, so the design has to be able to take on new capabilities without being rebuilt around them. Second, these years may bring sudden disruption: institutions failing, and sudden waves of people looking for exactly this. If a million people arrive at once, the pattern should still hold. Circles of seven grow by dividing, not by getting bigger.
Most platforms have to start by acquiring users. ServiceSpace starts with relationships built over decades, without advertising or fundraising.
Reflections, stories, circles, and offerings. With permission, these can help show where connection and service want to happen, without becoming something to mine.
The people building the tools have sat in the circles and done the service. They design from that experience, not from engagement metrics.
Because the work is gift-based and non-commercial, it can build tools that make people less dependent on the tools. A platform funded by extraction can't really do that.
When people help from genuine care, the helping tends to restore them. A good setup doesn't need to manufacture motivation. It just makes it easy for care to go somewhere.
Before AI, this kind of many-to-many coordination cost too much. Now it doesn't. And norms are still forming, so there's a window before extraction becomes the default.
All we can do is look after the conditions.
The age of AI won't only be shaped by bigger models, faster chips, and smarter agents. It will also be shaped by whether people can still gather in trust, listen across differences, help each other without keeping score, and remember that love is a kind of intelligence too.
AI may out-think us.
It can't love for us.
That part is ours to practise.
A starling doesn't watch the whole flock. It watches the few birds nearest to it, and the flock turns together.