On My Radar – Week of May 18, 2026

Anthropic's reference piece on context engineering — what to put in the window, what to leave out, and why. The canonical framing for the topic.

Rich Sutton's classic essay: general methods that scale with compute beat hand-engineered knowledge. The lens behind most agent-harness design debates.

Vercel on why fewer, better tools beat a sprawling toolset — fewer tokens, less confusion, better task completion. Contrarian and concrete, directly relevant to CX agent design.

Chroma's empirical study showing model performance degrades as input context grows, even well below the window limit. Heavily cited in the context-engineering debate.