Agents
From RAG to the Library: History Rhymes
RAG and skills aren’t new ideas — they mirror how humans have always retrieved knowledge. History suggests the destination: the library, with its hierarchical organization built for navigation at scale.
Agents are features not products
The pitch is that AI-first companies don’t need traditional tooling, that known approaches are dead. The nuance: models can only one-shot without an intermediate piece to iterate on, and that piece is what traditional approaches produce. The model multiplies the operator; it doesn’t replace the workflow. Just like GPS software vanished into car consoles, agents will vanish into products.
Smart Agents Don't Need Managers
The irony of ‘PhD-level’ agents needing human orchestrators is fading. Context windows and on-demand subagents make manager layers obsolete.
RAGs, MCPs and Scripts. AI moves fast
How traditional RAG is being rapidly replaced by agent-based querying and lazy-loading systems like Claude Skills, and why scripting provides superior context window management