She's both, right and wrong. Right because lazy agentic implementations will do that and people will fall for it.
Wrong because there is nothing inherent about AI agents that say it has to be implemented this way.
Every operation she mentioned can be implemented in isolation in a zero knowledge fashion. AI doesn't need access to "your browser". It needs access to "a browser", one with no cookies, in a sandbox on VPN, and recycled after use. All of the other operations can be done by your own private, local-running MCP servers, so the AI knows nothing but what the tool is saying.
Finally, the convergence of better, more intelligent models with less parameters and the quality of new hardware will make local LLMs flourish.
neo.metal /
npub146…lpzrj
2025-03-10 19:48:23
in reply to nevent1q…ql9u
Author Public Key
npub146p345u4hxxxc2ugzthvhyrus076zdj7dmlhnqcwk36l4xqt64ysdlpzrjSeen on
wss://relay.nostr.bandPublished at
2025-03-10 19:48:23Event JSON
{
"id": "f66efd9ef6cb0f11aec008ba37a3f05d351638e5acb666370bc702763fd6c825",
"pubkey": "ae831ad395b98c6c2b8812eecb907c83fda1365e6eff79830eb475fa980bd549",
"created_at": 1741636103,
"kind": 1,
"tags": [
[
"e",
"b82aad5f633e92840daf3034fd0dc93bbbc23940ed160188719eb48a3725db11",
"",
"root"
],
[
"p",
"472f440f29ef996e92a186b8d320ff180c855903882e59d50de1b8bd5669301e"
],
[
"r",
"wss://nostr.land/"
],
[
"r",
"wss://premium.primal.net/"
],
[
"r",
"wss://relay.damus.io/"
]
],
"content": "She's both, right and wrong. Right because lazy agentic implementations will do that and people will fall for it.\n\nWrong because there is nothing inherent about AI agents that say it has to be implemented this way. \n\nEvery operation she mentioned can be implemented in isolation in a zero knowledge fashion. AI doesn't need access to \"your browser\". It needs access to \"a browser\", one with no cookies, in a sandbox on VPN, and recycled after use. All of the other operations can be done by your own private, local-running MCP servers, so the AI knows nothing but what the tool is saying.\n\nFinally, the convergence of better, more intelligent models with less parameters and the quality of new hardware will make local LLMs flourish. ",
"sig": "b385115dbb46f2b6228f64c0dee202c6a1ec089471509856c78a3efeb047faef32abd8807824ecf9620429cfd56619d221d32a28f9dbb9c8162b41328d2556f1"
}