This paper proposes a file-system abstraction for context engineering, inspired by the Unix notion that 'everything is a file'. The abstraction offers a persistent, governed infrastructure for managing heterogeneous context artefacts through uniform mounting, metadata, and access control.
It's sort of like Claude Agent Skills but I feel dexo better, I saw some agent use MCP server as backend and unlike Agent Skills install on the client.
The combination of user ownership (decentralization), quantified quality relationships (Social Tokens), and simplicity (AI Agents) fundamentally shifts the incentive structure. We move from an environment designed to extract attention to one designed to build verifiable, lasting social capital.
The new social contract is clear: You are the owner of your digital home, and your true value is measured not by how many strangers click your ads, but by the quality of the relationships you invest in.
Great point — I actually agree with your long-term view. Autonomous AI agents acting as “full pilots” is a compelling direction, especially for clearly defined tasks. But I’d add that HUDs and copilots aren’t mutually exclusive — they often coexist, just like in real aviation.
In fact, the HUD metaphor comes directly from the cockpit: it enhances situational awareness while the pilot (human or AI) focuses on action and decision-making. HUDs serve a different role — not replacing pilots, but grounding them with continuous, low-friction perception.
Not a promo — just sharing our design rationale for ArcSphere and why we went with an AI HUD over a Copilot model.
I recently read the article “Enough AI copilots! We need AI HUDs,” and it resonated deeply with me. It perfectly echoes the design principles I’ve been exploring for months with ArcSphere, our AI-native browser.
