A working bench for the AI side of the practice — evals, product thinking, workflow agents, MCP and automation. Some bits are live, the rest are next.
Score real prompts against a real rubric. Browse the iterations behind the article-insights panel, or run your own and grade the output.
How I scope, validate and ship AI features without falling into demo-ware — picking problems where the model is the cheapest path to a useful answer.
Patterns for chaining tools, retries and human checkpoints — what works in practice once you stop treating agents as magic.
What this site exposes over the Model Context Protocol — the tools any compatible client can discover and call, plus how to connect.
Concrete automations that paid back the build effort — what was worth doing, what wasn't, and why.