Mylobster.ai Plans OS-Level AI Assistant

- Mylobster.ai founder Dendi Suhubdy argues that current AI assistants are limited by "userspace" access, relying on APIs and integration tokens, whereas a kernel-level position provides privileged visibility into system calls, file operations, and network traffic. - The technical approach for this deeper integration involves loadable kernel modules and eBPF programs on Linux, and kernel-mode drivers using KMDF with filesystem and network filtering on Windows. - Integrating AI at the kernel level introduces significant security and privacy concerns, as it could expand the attack surface for adversaries and bypass user-space safeguards like application sandboxes. Malicious actors could potentially inject biases or backdoors into AI models during training, which would then manifest as kernel-level exploits. - The push for OS-level assistants is influenced by the rapid adoption of open-source personal agents like OpenClaw, demonstrating a demand for automation that can perform actions directly on a user's machine. - This architectural shift raises fundamental questions about authorship and agency in creative work, moving the AI from a simple tool to a "co-creator" that can suggest novel concepts and streamline complex design tasks. - In practice, effective human-AI creative workflows often break down tasks into sub-activities, with varying degrees of human and AI participation in idea generation, evaluation, and refinement, a model used by Netflix for script development. - The concept of an "AI Agent Operating System" (AIOS) is emerging as a framework to manage memory, tool execution, and inter-agent communication, treating AI agents like applications within a larger ecosystem. - Mylobster.ai views the personal assistant market, currently valued around $15 billion, as a stepping stone to a much larger $200 billion opportunity that includes IT support, productivity software, and endpoint security.