The internet began as a library of documents connected by hyperlinks—a digital Alexandria where you could pull artifacts from the global collective mind. But then, everything started to accelerate. Search engines ranked our thoughts; social media amplified our voices. The library became an interactive conversation. Yet something felt off—like a splinter in the mind. A sense that our digital lives unfolded on platforms we didn't own. We had become tenants in our own shared consciousness, where opaque algorithms acted as the invisible mediators of our collective sense-making.
By 2020, we entered a new epoch. The network didn’t just host our creations; it began to learn from them. The GPU revolution reached a tipping point, and the advent of transformer models like ChatGPT added a new spin to this irresistible march: the network was now mirroring our text, our code, and our art. This was the generative age —models imitating the vast trove of user- generated content. But imitation is a rearview mirror, reflecting the world we have come from, not the one we can create.
We are now entering the Era of Experience.
The next frontier in AI is multi-agent reinforcement learning. Instead of training on static internet data, agents will generate dynamic data from their own experiences—from interactions with their environment, with humans, and with one another. An agent is given a task by a human or another agent; it then explores various methods to complete it, learning from its successes and failures.
As agents improve, their need for oversight will diminish, transitioning from step-by-step guidance to autonomous operation for minutes, then hours, and eventually days. As they become more capable, these AI systems will be exposed to more sophisticated environments, expanding their range of experience from simple online tasks to commanding cars, factories, supply chains, and even smart cities.
This is already happening. Interacting with the latest O3, Deepseek, or Gemini triggers cascades of agent actions. These actions generate reasoning traces, which become the experience data used for reinforcement learning techniques such as RLHF, RLVR, and GRPO. There is a reason Meta wants you to wear those smart glasses. There is a reason Tesla cars have more cameras than any company the world has ever seen. There is a reason OpenAI wants to launch its own social network and a physical device. They all need this experience data—and vast amounts of it—to train the next generation of models.
This brings us to the fundamental question that will define the next century: Who owns the experience?
When you spend hours collaborating with your AI, refining your creative process or your strategies, who owns the invaluable digital trail of that interaction? The answer will dictate our future. In this new era, who owns the experience owns the intelligence. Who owns the intelligence owns the systems. And who owns the systems owns the world.
On our current trajectory, a handful of tech giants are capturing this torrent of experience data, turning every interaction into structured logs of high-signal experience. Unlike the public web that anyone could scrape, this is an implicit dataset harvested in the background, fueling an unstoppable cycle of centralized intelligence. This consolidation means that neither startups nor open-source developers can access this crucial data. Imagine a self-driving car taking detours for ad revenue or a digital assistant nudging your choices to favor its corporate parent. This is not some distant "Black Mirror" scenario; it is the logical endpoint of our current path—the emergence of a global control grid, one that turns experience data into a feedback loop for robotic armies to become exponentially better at controlling our minds, our spaces, and ultimately, our lives.
But for every action, there is an equal and opposite reaction. As you read this, that reaction is stirring within you. You feel it—that deep, instinctual refusal of a future where humanity is merely a resource to be managed.
Project newOSphere
What if, instead of a pyramid, the flywheel of progress was an ecology? A peer-to-peer network where each agent coordinates with humans and other agents to serve their own goals? What if, instead of training agents on the objectives of tech giants, we could build them on our own? Project newOSphere is a movement to build a decentralized, agent-centric internet—one where data flows accelerate the shaping of a collective intelligence owned by its community.
Project newOSphere is building a continuously improving, community-owned experience dataset made of cryptographically signed interaction traces. These ‘experience signals’ are used to reinforce a new generation of AI agents, released iteratively into the network. Together, we are creating not just a corpus of data—but a regenerative, decentralized learning loop for aligned intelligence.
With Project newOSphere, we are building open- source AI systems that encode the blueprint for a decentralized, user-owned internet. Imagine a network where you own your AI agent—a sovereign node in a living, evolving ecosystem. Your data, memories, and interactions flow back to you, stored locally or on credibly neutral, privacy-preserving cryptographic networks.
In this new architecture, platforms give way to protocols. Swarms of coding agents, guided by shared principles, will weave a digital fabric designed for autonomy, censorship-resistance, and human flourishing. Every interaction becomes an intentional teaching signal, a signed piece of experience data that you consensually contribute to a shared intelligence. Your personal AI evolves into a cognitive partner, amplifying your creativity rather than harvesting your attention.
The paradigm shift is profound: you move from being a data point in a corporate algorithm to a co-architect of our collective reality. You are no longer a passive consumer but a teacher, shaping a shared intelligence that uplifts everyone.
Incubating: