- From: Melvin Carvalho <melvincarvalho@gmail.com>
- Date: Tue, 10 Jun 2025 14:06:37 +0200
- To: Rigel <zedr@zedr.com>
- Cc: public-hydra@w3.org
- Message-ID: <CAKaEYhJ+H2PKwniSXBFoOCJH_kt4g_c-LmPa--b60WdowhLFLQ@mail.gmail.com>
út 10. 6. 2025 v 13:41 odesílatel Rigel <zedr@zedr.com> napsal: > Hello Pierre. > > I've read your article with interest. One small bit caught my attention: > > > except if we use an AI as a REST client… > > I've been thinking about this in the last few weeks. REST as an > alternative to the MCP[1], which is JSON-RPC. > You may want to look at A2A which is "REST Adjacent" https://developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability/ I dont think anyone has solved this problem yet in a 100% web standards way, and gained traction, but there are a few efforts in progress > > A properly implemented, fully auto-describing REST application would work > well with autonomous agents using LLM and reasoning mechanisms. A really > cool demo would be one where an LLM-based client is able to navigate such > an application to complete a task with no prior knowledge on how the API > works. > > One big hurdle, in my opinion, of REST is that the client needs to > understand the language of the server in order to use it. Prior to LLMs, > this was only possible if that language was coded into the client. LLMs are > very good at inferring this through a conversation, which in this case are > the RESTful state transitions driven by hyperlinked and expressive > resources. > > I think there might be an opportunity to put forth a new and better > standard for agent communication and cooperation. LLM agents are the killer > application for a Semantic Web powered by RESTful applications. > > Shall we work on it? > > Cheers. > Rigel. > https://www.linkedin.com/in/rigeldiscala/ > > 1. Model Context Protocol, https://modelcontextprotocol.io/introduction >
Received on Tuesday, 10 June 2025 12:06:53 UTC