What even is an OS now?
Advances in large language models are reviving an old dream: that ordinary users could “program” their own devices in natural language, raising questions about what an operating system should be in an AI-first world. Commenters weigh a vision of personalized, LLM-built micro‑apps and agentic assistants against the enduring need for traditional OS roles like resource management, isolation, security, and deterministic tools such as banking apps or spreadsheets. Many are skeptical that most people will ever design their own software, arguing that apps, GUIs, and curated platforms will persist even as AI becomes a powerful new layer on top of existing operating systems rather than a replacement for them.
What is an OS now?
- Several argue nothing fundamental has changed: an OS still manages hardware resources and isolates programs.
- Others note modern OSs sit atop opaque SoCs and firmware (e.g., management engines), so they no longer truly “own” the hardware.
- Some distinguish low-level kernels from the higher-level “OS experience” (shell, services, app model), saying the article mostly questions the latter.
AI, “vibe-coded” apps, and future of software
- Enthusiasts think LLMs enable many tiny, highly personal apps, written in natural language, changing the role of traditional app stores.
- Counterpoint: why build custom apps if you can just ask an AI assistant to perform tasks directly? Many expect a dominant general assistant instead of a galaxy of micro‑apps.
- Others foresee a hybrid: stable foundational libraries and services (codecs, storage, DBs) plus fluid, AI-synthesized tools or UIs on top.
Determinism, reliability, and safety
- Long subthread on deterministic apps vs nondeterministic AI:
- Pro-AI side claims nondeterminism is acceptable or even desirable for many everyday tasks.
- Skeptics stress engineering, compliance, and safety require deterministic, verifiable behavior; LLM-generated code is hard to trust.
- Some imagine AI generating deterministic tools behind the scenes, but emphasize rigorous verification and clear boundaries.
User behavior and power‑user bias
- Multiple posters say the vision overestimates typical users’ willingness or ability to “vibe‑code”; most just want stable apps and passive consumption.
- Historical parallels (BASIC, HyperCard, early web) are cited: a minority created; most did not, even though tools existed.
- Examples of people trying to get non‑technical friends/family to use LLMs for automation largely fail due to difficulty specifying needs.
Security, sandboxes, and platform control
- Strong concern that AI‑generated software heightens the need for strict sandboxing, capability systems, and permission models.
- Others predict more locked‑down devices and signed/“certified” software, justified as safety/security against unvetted AI code.
- Some worry AI-focused platforms could weaken user ownership and superuser control, especially on phones.
Economics, APIs, and the web
- If agents scrape sites instead of humans visiting, ad‑funded services may collapse; posters anticipate more paid APIs and machine‑to‑machine billing.
- There is skepticism about stacking costs (ISP + AI provider + data APIs) for simple tasks like checking the weather.
Meta reactions to the post and project
- Mixed reception: some find the vision exciting and timely; others see it as AI hype, vague on details, or a pitch for yet another phone.
- Several ask how such a system would coexist with existing regulated apps (banks, medical, social platforms) that demand strong OS‑level guarantees.