My PhD research proposed a web where software agents discover and use capabilities on their own — a decade before the Model Context Protocol made it real. This hub collects that whole arc: the prediction, the correction, and the blueprint for making agentic systems safe enough to trust.
Start with these
- From DeepGraphs to MCP — how the research anticipated the agentic web, and what it got wrong about the engine.
- Upgrading the Model Context Protocol — the MCP 2.0 blueprint: vocabularies, state machines, and neuro-symbolic guardrails.
- The 100x Agent Illusion — deterministic rails beat faster agents.
- The Open Source Loophole — the legal layer nobody updated for agents.
The lineage (2015–2018): the semantic web era
The agentic web didn’t appear from nowhere — it’s the vision the semantic web community chased for twenty years, finally given a working engine. These archive pieces are the receipts:
- Limitations of the Semantic Web — why the first engine stalled.
- Defining Digital Affordances, Defining the Digital Object and Applying Affordances — the DeepGraphs research trilogy.
- APIs and Linked Data · The Semantic Web: reality vs hype · The Versions of the Web
- Uniform Interfaces · A Hypermedia API Reading List · A Wikipedia for Web APIs
- HTTP/2 and gRPC — the protocol lineage that leads to typed machine contracts.
- Asimov’s Three Rules for the Web of Things — the very first post in this thread, from 2015.
Where it’s heading
Agents that plan, execute, and iterate are the first step of a much longer program: automating the mundane so humanity can aim at bigger things. That program is The APILama Tech Manifesto.
Related hubs: AI Strategy · Engineering Leadership