Towards Shared Minds

What started as an AI-native OS for our lab quickly turned into an internal obsession. Shortly after, we knew this was what we needed to build — and build it through it.

We are excited to share what culminated since then and some thoughts on why we think this matters for the present-future and a few mental models which we developed to think about this paradigm.

As Alan Kay put it, “the computer is an instrument whose music is ideas” — and the interface determines which ideas are playable. The terminal makes AI composable in a way no other interface does. Egregore turns running AI in your terminal (currently Claude Code) into a multiplayer experience by using a shared memory system and specific commands which facilitate coordination workflows. This base capability has a few key implications:

• context generation becomes a byproduct of organizational flows
• workspace and coordination space are unified in the most AI-native form factor, massively reducing switching and transaction costs
• organizational learning and adaptability which yields emergent capabilities downstream of needs

Mental Models

Since this is all genuinely new territory, let me share a few mental models for those who want to experiment with this. First off, Egregore is a relational context engine which proactively engages with your processes. Let it ask questions — presets for user responses are contextually generated but typing (or even better, speaking) your response will always yield better results.

I refer to this as context gardening (as opposed to engineering): the overhead approaches zero as context surfaces are pegged to the edges of the organization, and these edges multiply as the organization interacts.

Second, Egregore, with Claude Code, lives and operates on your file system. Make sure you sandbox the specific repositories you want to share with fellow egregorians. It is the living infrastructure of your group. Both its body and its soul will change through use. Lean into this and find ways to inscribe yourselves into the substrate for enhanced bi-directional sensing.

Egregore uses commands like /activity, /handoff, /save and /reflect to crystallize specific workflows to reduce cognitive load (both tokens and neuroelectric) of coordinated work. They are heteronoetic schelling points — crystallized for speed and cohesion, but designed to be mutated and composed for variety and evolution.

Beyond Single-Player

AI massively empowers solo users, however, it is clear to us — after a month building Egregore with Egregore — that groups who share context and frontier capabilities will be the most significant force in AI-native production. Today's AI discourse overfits everything to its priors debating whether your company is a file system or not, while the consequences of this for AI-induced organizational cybernetics are much wider.

Beyond the determinacy of enterprise or startup logic, such pluri-cognitive substrate could have profound political impact by facilitating social choice aggregation with unprecedented granularity, constituting very credible evolutive routes for “democracy”. Egregoric substrates could become primary media through which open-source agentic capabilities are ported and shared, selected through generations of experimentation. The path that Hugging Face charted for OS models will be extended to agentic development and learning by Egregore. With radical generalizability, solo-entrepreneurs with many-agents can use their Egregores as a compounding memory and coordination surface utilising the cutting edge practices from the community.

Egregore is an organizational substrate of co-learning and selection — an environment whose primary output isn't decisions or products but an expanding variety of possible configurations, coordination patterns, and capabilities. The wider the gene pool such organizational space maintains, the more of the future it can metabolise.

A Note of Sobriety

Having said that, it is worth microdosing a bit of sobriety along with the utopian delirium. Egregore works best with daily organic usage across work and research contexts. The ability to work with the frontier model in Claude Code through long sessions is currently a privilege few can afford. These costs, both on the model side and Egregore optimizations, are bound to come down rather quickly — but we need to be frank that being an Egregore power user is currently most suitable for Claude Max users (which costs >100$/m). For instance while building Egregore, daily >100$ bills from Anthropic weren't uncommon.

Besides the costs, the other challenge we should face is cognitive security. AI is already a pitch-dark forest, with adversarial techniques popping up as fast as new capabilities. As interfacing with AI becomes a collective enterprise, the attack surface increases together with the network effects. This is something we need to engage with proactively, hopefully as a community of builders.

Origin Story

Before Egregore had a name, Oguzhan (Oz) and I had been experimenting for almost a year with dynamic knowledge graphs and emergent ontologies. Things we built somehow lacked a critical piece, which was hard to see at first. At some point we decided to radically recalibrate — to stop thinking about product, and to remodel how we think and work from scratch, without overly attaching to any form factor. Simply play – together.

A few days later, at a cafe in Kreuzberg, we were deploying our AI-native “lab OS”, unsure where it would take us. Within a week we were exchanging handoffs every day, asking for context across workstreams, logging research. The magic hit fast. Every time we did something outside of Egregore we started questioning why — why pass on the opportunity to generate collective context?

What followed was an insanely animated month for me, Oguzhan (Oz) and the Curve Labs team, obsessively dogfooding and shipping features. We can't wait to finally push this out, whether that means validating our visions or delusions.

We are starting with an alpha testing phase through a waitlist. We will engage with design partners who are capable builders, open to sharing their experiences and genuinely excited about what Egregore can become and what they can do with it. Hopefully in the near future we will get to a public release.

Til next time.