Zulip, MCP, and AI Infrastructure for Organizations
No one is talking about the most important decision for leveraging AI across an org: the communication infrastructure that joins teams of humans and teams of AI.
At WindBorne Systems, AI agents are deeply embedded in how we work and communicate: they're all over our chat app, Zulip. Individual AI use doesn't scale to most org functions—you need one shared nexus for humans and bots to read context from. This isn't possible on closed-source apps like Slack, where you can't build custom integrations without getting rate limited or waiting on app-store approval.
Good infrastructure, modifiable open source software, and a leadership team full of engineers unlocks God Mode. Some examples:
- Our growth team (completely non-technical) runs WeatherMesh case studies by pinging an evaluator agent in Zulip. Done on their own, this would lead to analysis errors, but engineers can see the conversations and correct the tools and teach fellow humans.
- Our deep learning team has agents starting and monitoring all training runs—many barely look at code anymore. If a run fails, bots ping on Zulip while they debug, posting as they go.
- I vibecode on Zulip while walking through the grocery store.
- We have AI auditor agents reviewing our codebases, datasets, balloon flights, and deployed software for issues, bugs, and security vulnerabilities, sending reports on Zulip.
The Infrastructure
This isn't trivial. It requires a LOT of infrastructure investment. Some examples:
- A Zulip MCP server that lets agents read and write messages, which I'm now open-sourcing: windborne/zulipmcp. I've rewritten this three times over the last year, and finally have something I feel may be worth sharing.
- A custom Zulip search engine (with MCP) that embeds 350K+ conversation chunks into a vector database.
- LossVegas, a custom W&B replacement I vibecoded 9 months ago, which agents now query autonomously via its built-in MCP.
- Dozens of other custom MCP servers for internal tools, databases, and checks.
- One of our most senior engineers, Davy Ragland, moved from running Atlas Software to 100% focus on building our agentic systems.
Of course, we bootstrap: the agents build their own tools with our guidance, but it's tricky to get everything right.
Looking Ahead
Over the next 12 months we'll see a bifurcation between companies willing and able to go all in on the singularity, and those that cannot. You HAVE to reengineer your organization yourself—you cannot just buy a SaaS product that leverages AI for you.
Not every company can do this. I'm fortunate to have such an incredible team at WindBorne, capable of learning these new methods and building for ourselves. AI will let us stay small, fast, and efficient while building this Planetary Nervous System spanning all of Earth.
