Subconscious is winding down
Subconscious started with an idea: to amplify intelligence using a worldwide decentralized knowledge graph. In 2022, we raised a seed and started building.
Today we’re sharing difficult news. Subconscious will be winding down active operations. I’m proud of what our team built on a technical level—a multiplayer graph protocol for notes. Despite this, we ran into difficult technical headwinds and missed a crucial market window. Ultimately, we did not find product-market fit.
I now believe we were aiming at the wrong target. The world changed in a fundamental way while we were building. Today, if you want to amplify intelligence, you probably wouldn’t build a decentralized notes graph, you would go to work on personal AI. Subconscious, though, isn’t the right vehicle to pursue this scale of opportunity. The shape is wrong, the runway too short, so we’re calling the ball.
That’s the hard news. But if you ask me, the dream of a bicycle for the mind, of software that amplifies intelligence, creativity, and agency, is more real than ever. AI has unlocked a wild new landscape of possibility. There is no time to lose. I’m venturing forward to build something new. I’ll keep you updated here on the journey.
To our beta testers
The Subconscious Beta TestFlight will expire after a few months. However, you can export all of your Subconscious data as useful plain text files, through the iOS file system and the Noosphere protocol.
Guide: exporting your Subconscious data.
Thank you for testing and offering feedback on the Subconscious Beta. Your feedback has been invaluable. We appreciate how thoughtful and kind you have been, and that you’ve taken the time and energy to test-drive early software.
This newsletter will continue
This newsletter will become my personal lab notebook. I’ll continue to post here, but under my own name, as I navigate the idea maze around AI, decentralized protocols, and tools for thought.
The Internet was done so well that most people think of it as a natural resource like the Pacific Ocean, rather than something that was man-made. When was the last time a technology with a scale like that was so error-free? The Web, in comparison, is a joke. The Web was done by amateurs. (Alan Kay)
My motivating question is: how can we build more things like the internet? Or just grant me the amateurism to create things like the web! I would like to build software like this—software that is open-ended and evolvable, software that amplifies our agency.
What I think we got right
When we started building, I had a strong intuition that a moment of opportunity was coming to decentralize the social web. Here, I think my intuition was right.
My hypothesis was: make a multiplayer tool for thought by combining something like Roam with something like Twitter.
Roam resurrects a vision of the web we never got—the JCR Licklider Memex Web, the Doug Engelbart Augment Web, the Ted Nelson Xanadu Web—hypertext, as it could have been.
Twitter, at its peak, functioned as a kind of exobrain.
Combine these with a decentralized protocol, and maybe it could be the seed of a new web-like thing?
This was an unusual pitch in 2021. But then, in April 2022, Elon Musk bought Twitter. By summer 2023, a wave of protocols was washing over the internet in reaction.
Farcaster launched a permissionless network
Bluesky surpassed over 2 million users
Threads implemented ActivityPub
The wildest part of our thesis was being vindicated by reality. The social web wanted to be protocols! Furthermore, these projects converged toward similar architectures, mostly using self-sovereign keys, decentralized data, and ordinary servers.
How is it that so many protocols were ready to take on Twitter, just months after Musk bought it? Why did we see convergent evolution around protocol architectures? Others had also sensed a shift in the Zeitgeist months in advance, and started building. These weren’t disconnected events. They were the result of structural forces playing out in a late-stage ecosystem.
The old web is dying, a new one struggles to be born.
Missing a window
While we were directionally correct, the unfortunate fact is that we missed a window while struggling through some difficult protocol integration challenges.
Hot Protocol Summer was a one-time event, a moment when the shock of Musk’s Twitter takeover enabled open competitors to bootstrap viable networks.
Ecosystems have overwhelming momentum. It’s only during these moments—when ecosystems collide, die, or are disrupted—that we have agency to tilt the trajectory of the future. These windows of opportunity are fleeting and contingent. They open up at the beginning and end of s-curves, after crashes, or during collisions between previously isolated ecosystems. Timing them correctly is crucial.
By Q1 2024, it was clear Farcaster, Bluesky, Nostr, and ActivityPub all found early product-market fit, and we hadn’t. We missed a critical window.
Chaos is a ladder. The ladder has been pulled up for the time being.
Aiming at the wrong target
Our goal with Subconscious was to amplify intelligence. The best way to do that in 2021 looked like building a multiplayer tool for thought. Combine something like Roam with something like Twitter, and if you manage to bootstrap a network, you might get collective intelligence.
Then ChatGPT launched, and within months, good-enough open source LLMs were everywhere. This upended our hypothesis.
Tools like Roam, Obsidian and Logseq are hired to amplify intelligence—that is the JTBD. So, but, if I’m looking to amplify my intelligence today, what am I going to reach for? Probably Claude. The basis of competition has shifted.
This makes me question many of my previous assumptions:
Instead of evergreen notes, what if we hallucinated notes on demand?
Instead of knowledge graphs, what if we generated embeddings?
Instead of a multiplayer, what if we had hundreds of friendly NPCs?
Do we need notes at all?
I don’t know? It’s a green field! LLMs change the way computers think, and the ways we can think with computers. Whole new classes of tools are possible.
In 2024, it is clear to me that if you want to amplify intelligence, you don’t build decentralized notes, you work on personal AI.
What’s next
LLMs are a big deal. They disrupt tools for thought, sure, but also the web, and culture at large. We’ve built much of our world around text. Now the text is speaking back. It has a mind of its own.
This changes things. Software is getting softer, and the dream of computers as personal dynamic media suddenly seems more tangible.
There’s lots more to say, and I would like to share more of what we learned while building Subconscious, but it will have to wait for a future post. For now, I’ll just say a heartfelt thank you for accompanying us on this journey.
Stay tuned.