✨🙌 AI that helps communities thrive on their own terms

Open protocols + AI-enabled coding = building what we need for ourselves

We’re thrilled to support the upcoming ATmosphere Conference, in Vancouver, Canada from March 26-29. Register to attend.

Last summer we welcomed Blaine Cook, Principal Software Engineer for Roundabout, our local community app. Blaine goes all the way to the beginning of Twitter, and he has been hard at work with the Roundabout team creating this brand new platform, which we’re now piloting in communities throughout the US.

We’re using AI in this work, informed by our AI Policy. We remain humble and alert to the risks posed by these tools, but also curious and open to experimentation. If we might be able to use new AI tools to bring an open, useful, human internet to more people, we’re going to try. That’s what Blaine’s piece is all about. What are we trying to do with these tools?

Reminder: send your plugs and requests for the next Bulletin Board.

– Josh Kramer, Head of Editorial, New_ Public

Within the larger movement to build better, public-spirited apps and websites for the social internet, the conversation around using AI usually defaults to a familiar set of applications. There is a safe, incremental approach to building prosocial software where we use large language models to address a limited set of features:

• Provide moderation and safety

• Improve ranking and sorting

• Help with content management

We should do all of these things, but this framing reflects a deeper assumption inherited from large social media platforms — that the primary role of AI is to extract value from centralized data and optimize what is shown to users.

That assumption is worth challenging, because these tweaks are not the most exciting possibility for AI and the social internet.

At New_ Public, and in the design of Roundabout specifically, we start from a different thesis: social value does not emerge from technology. It emerges from people — real, local, invested participants who curate, organize, and care for their communities. The risk of the dominant AI framing is not just that it recentralizes power, but that it actively disempowers users by treating social life as something to be optimized for them rather than shaped by them.

The more transformative opportunity is to give more people the ability to do what early adopters are already doing today: not AI as an invisible hand guiding social life, but AI as a tool to help restore agency to communities themselves. Instead of legacy fixes that preserve the status quo, we should be encouraging and supporting each other to build tools for our own communities, like our friends and peers at Blacksky are doing on AT Protocol.

While this remains inaccessible to many in the short term, there’s a clear trajectory. We should provide robust, reliable, and trustworthy core infrastructure, in the form of open services built on and supporting decentralized protocols. At the same time, we should radically expand who gets to participate in shaping this ecosystem. That is a future worth running towards and fighting for.

LLMs Should Not Curate Our Social Lives

Large language models are extremely good at “data munging”: transforming text, summarizing content, extracting structure, and accelerating software development. That’s why it’s no surprise that the dominant platforms have been eager to insert these models into existing moderation and ranking systems. However, they are less good at having taste.

One of Roundabout’s core principles in using AI is that nothing should reach the site without meaningful human involvement and decision-making. This is an ethical stance, supported by New_ Public’s AI Policy, but we’ve also found that it leads to higher quality and more prosocial outcomes. I often think back to a sign at a friend’s office: “If we betray your trust by giving you inferior candy our loss will be greater by far than yours.”

Take event curation as an example. Using today’s LLMs, it’s easy to automatically search the web, find events that are happening in a community, and compile them into an event calendar. With Roundabout, we’ve tried experiments to this effect, but they consistently miss the nuance, taste, and trust signals that can only come from the people who make a community.

Indeed, it’s acts like organizing, debating, and celebrating that often creates community. Even if we have the good intentions to automate and remove genuinely tedious administrative work, we must be relentlessly focused on which processes we replace and why. Making communal acts of mutual devotion obsolete is a soulless stripping of our humanity, and the opposite of what we’re trying to do with Roundabout.

This distinction is critical. LLMs shouldn’t decide what we see, but they can help us build the tools we need to support our communities.

Image courtesy of Flickr user Matt Biddulph, via Creative Commons license

The Real Opportunity: LLMs as Tool-Builders, Not Social Arbitrators

I’ve spent a lot of time with Claude and other AI coding tools over the past year, and talking to other experienced practitioners who have done the same. I think it’s fair to say that LLM-assisted coding has dramatically lowered the barrier to creating bespoke software. The consensus is that building simple one-off tools is now trivial; a little experimental app rarely takes longer than fifteen minutes in the right hands. That said, it’s also true that building high-quality and reliable software for wide audiences will remain a significant undertaking.

Even though code is now effectively free, polish and depth takes time. Our team has spent thousands of Claude-assisted hours over the past ten months building Roundabout. The hard part, as it has always been, is deciding what to build and making it good.

Even if this pace accelerates to the speed of thought or beyond, Jevons Paradox applies to software, too. The backlog of features and integrations any social platform might want to build is effectively infinite. No centralized team — even one moving quickly with AI — can keep up.

We’re still early days with this line of thinking, but we’re doubling down on this at Roundabout. Rather than building a product that accretes and monetizes all of the value and knowledge of the people who use it, we want Roundabout to help foster the creation of robust, shared prosocial infrastructure that community members can build on top of and use to address their own needs, on their own terms.

For now, this looks like calendars, guides, and other features that community stewards and members can shape for their own use. But this is just the beginning — we believe that combining open infrastructure with the enabling aspects of LLMs could help unlock the creativity inherent in communities.

One example that I personally hope to see is that a group helping with emergency coordination, like Watch Duty, could build tools that integrate directly with Roundabout. Having personally experienced a major wildfire, I know firsthand how important these tools are, how complex and context-specific they can be, and how often they are only available for the highest-profile disasters.

Roundabout as an API for Communities

It might be helpful to think in terms of another piece of digital infrastructure — APIs. Application Programming Interfaces are protocols that shape how data flows through the social internet, but there are important decisions to make about who and what controls that flow.

Facebook once offered APIs, so third parties could build on top of Facebook data, but they were fundamentally extractive. Control and consent ultimately resided with the platform, not with users, allowing Cambridge Analytica to harvest user data on a massive scale — a predictable outcome of centralized power paired with opaque data access.

Decentralized networks like the AT Protocol, which powers Bluesky and also underpins Roundabout, create the possibility of flipping this dynamic. When combined with AI tools, consentful protocols — where users control the flow of data — could enable a new model. Users could create tools that interact with their data, compose services across systems, and, if necessary, revoke access when trust is broken.

As a small example, our Roundabout community stewards were requesting a post scheduling feature. We quickly built that with LLMs, and we’ve added that open source code back into the AT Protocol ecosystem, and made it available for the 41+ million users of Bluesky. It can be modularly adopted by any user, or instantly disconnected.

If Roundabout prioritizes consent as a requirement for accessing community data and interactions, then it would become possible for end users — not just professional developers — to safely create bespoke tools for many aspects of civic and social life:

• Event aggregation and promotion

• Community-specific moderation workflows

• Governance tools

• Local advertising and sponsorship experiments

• Political and civic organizing

• Emergency management tools

• News and reporting tools

• Analytics that communities control and understand

The audience for Roundabout, then, is not only end users, but a potential ecosystem of tool-builders embedded within communities themselves. These folks might not think of themselves as developers, but the neighbor who organizes a block party knows more about their own needs than we do, let alone any number of teams at Silicon Valley tech startups. (This is the premise of what our friends at the Relational Tech Project are working on.) Given the right tools and agency, we’re sure the things they’ll build will be remarkable.

My long-standing hope has been that we’re able to move past the extractive, monopolizing, and competitive phase of social networks, and into a new era of creativity, collaboration, and diversity. I believe we’re poised to see a Cambrian explosion of new ways to interact online, and there’s evidence to suggest that it’s already happening: just today, I saw three new apps to share what you’re reading and watching with friends, each with their own unique take on the subject!

In this light, LLMs may be a killer app for decentralized networks — and decentralized networks may be the missing constraint that makes LLM integrations safer, more legible, and more aligned with user interests. It’s a symbiosis, and I believe we need both pieces. Rather than trying to integrate LLMs with everything, I think that deliberately bounded, consent-driven integrations will produce better outcomes.

None of this lets AI companies off the hook for the harms they have and will inflict — at the same time, our ability to build more flourishing digital public spaces depends on the strength of our networks and our communities.

In contrast to so many enshittified companies with rallying cries of “aut Caesar aut nihil,” at Roundabout we believe in a resonant internet, where we work with our peers to support and enable people in our communities. A lot more is possible if we build it together.

– Blaine Cook, Principal Software Engineer, Roundabout

---

Thanks Blaine!

Looking forward to the cherry blossoms, azaleas, and rhododendrons,

–Josh