Where things stand

One person, a swarm of AI agents, and a bet on the end of the year.

I've been building Adama since 2020. It started as a board game server, turned into a programming language, then a runtime, a storage engine, a networking protocol, and a reactive HTML framework. That's too much for one person. I know that. I've always known that. But here I am, still building it, because the core idea -- that your backend should be a document, not a pile of services -- keeps proving itself right every time I use it.

So let me be honest about where this is heading.

The platform is good. I can't support it.

Adama works. I use it. The living document model, the reactive types, the delta sync, the privacy-as-a-language-feature thing -- all of it works. I've built real applications on it and the experience is genuinely great. Write a text file, deploy it, get a backend that handles persistence, real-time sync, privacy, and durable workflows. No stitching together six services. No cache invalidation bugs at 2am.

But I'm one person. I don't want to hire. I have a litany of reasons for that -- some philosophical, some practical, all of them deeply felt -- and I'm not going to pretend otherwise. Running an infrastructure company means on-call rotations, support tickets, SLAs, and all the organizational gravity that comes with promising other people uptime. I've done that life. I hated it. I'm not going back.

Which means the "Adama Platform as a hosted service for you to build on" story is dead. I'm keeping the platform in a personal monorepo, and the public face is adama-core: the language, the runtime, the solo server, MIT licensed. If you want it, it's yours.

The AI pivot is real

My entire strategy has shifted towards AI. Not in the "sprinkle some LLM calls on top" sense. I mean structurally. Agents are becoming first-class citizens in the Adama language -- persistent, stateful, transactional, with compiler-enforced token budgets and tool boundaries. An agent in Adama isn't a bolt-on framework; it's a reactive type that participates in the document's commit cycle, survives crashes, and delta-syncs to clients like everything else.

And I'm building with AI too. I'm a one-man crew leveraging a swarm of AI agents to help me ship. The irony isn't lost on me -- building agentic infrastructure with the help of agents. But it works. The code gets written, the tests pass, the ideas move forward faster than they have any right to for a solo project.

The end goal: agents and people can leverage Adama as a solo service. Your document is the server, your agents live inside it, and the whole thing runs on a single node with the economics of sleeping documents. That's the pitch for the end-of-year launch.

Products, not infrastructure

I've spent years shipping a meta-project. Language design, compiler work, storage engines, networking protocols -- all of it necessary, all of it invisible to anyone who isn't me. I've been shilling infrastructure to a world that doesn't care about infrastructure. The infrastructure has to manifest in a product that people can touch.

So that's the shift. I'm building a flagship product on Adama. Something real, something people use, something that demonstrates the model works in practice and not just in my head. The platform gets better because I'm using it, not because I'm endlessly polishing it for hypothetical customers. Dogfooding as strategy. Build the thing, use the thing, fix the thing when it breaks under the weight of the thing.

If I'm being honest, this is also how I trick myself into finishing. I have to maintain a handful of options for what I work on, and switching between "make the product better" and "make the platform better" keeps the momentum alive. The decisions are already made such that I have a lot of progress. Now I need to ship.

The bet

End of year. adama-core as a solo service that agents and people can run. Agentic support baked into the language. A flagship product that proves the model. One person, an army of AI collaborators, and a runtime that's been quietly getting better for six years.

Will I ship? I think so. But I've been wrong before (ask me about my previous timelines). The open question is whether the world needs another way to build backends, even one this opinionated. I believe it does, especially as AI agents need stateful, durable, privacy-aware environments to operate in. But belief and market reality are different animals.

The important thing is that I'm building something I love using it. That part isn't a bet. That part is already true.

How you can help

Star adama-core on GitHub. That's the least you can do, and it genuinely matters for a solo project competing for attention. If you want to go further, join the Discord -- the link is on the repo page. And follow me on X if you want updates. I'm building in the open, and I'd rather have a small group of people who actually care than a large audience that doesn't.