Blog
Field Notes
Things I’m working on, exploring, or paying attention to.
July 8, 2026By Roland CadavosContext engineering
Context Engineering: When the Prompt Stopped Being the Point (2026)
‘Prompt engineering’ grew up into something less glamorous and more useful: assembling the right information into a finite context window, and ruthlessly leaving the rest out.
Read post
June 18, 2026By Roland CadavosAI evaluation
Evals Are the New Tests: Measuring Agents Before You Trust Them (2026)
Shipping an agent without evals started to feel like shipping code without tests. The teams that stayed sane in 2026 turned real failures into graded datasets and wired them into CI.
Read post
May 14, 2026By Roland CadavosMCP & integrations
MCP: The Integration Layer Agents Were Missing (2026)
By mid-2026, the Model Context Protocol had quietly become the USB-C of AI tooling: one contract that lets any agent reach your data, tools, and services without a bespoke adapter for every client.
Read post
April 2, 2026By Roland CadavosOpenClaw
OpenClaw: Local Agents Meet the Real Developer Toolchain (2026)
OpenClaw pitches a different shape than chat-in-the-cloud: agents that run close to your machine, wire into Git and issue trackers, and treat skills and plugins as first-class—if you accept the ops cost of something that can actually execute.
Read post
March 27, 2026By Roland CadavosAnthropic
Anthropic: Claude, APIs, and the Safety-Shaped Product (2026)
Anthropic’s bet is that capable models plus disciplined API design and a visible stance on safety can win enterprise trust—while developers still care most about latency, context, and whether the tool fits their stack.
Read post