Custom Rubrics for Agentic Search
July 13, 2026Platform search fails niche users because it optimizes for the majority. Custom rubrics—private criteria applied by AI—let you find what you actually want by reading the signals businesses can't fake.
I write about how AI changes software, work, and search—and about what remains valuable as model capabilities improve. These essays form a path through that argument, from the big picture to concrete ways of working with models.
Start with Apps After Agents for the core thesis: models are replaceable, while the judgment that accumulates around them is the durable substrate.
Platform search fails niche users because it optimizes for the majority. Custom rubrics—private criteria applied by AI—let you find what you actually want by reading the signals businesses can't fake.
Every generation of programmers believed their ground was solid. Assembly, C, manual memory management — each felt essential until the boundary moved. Now AI is moving it again.
Models are commodity. Substrate is differentiation. The question isn't how smart your agents are — it's where the crystallized judgment lives.
LLM prompts are like flying cameras through latent space, capturing snapshots of frozen intelligence.
Building a YouTube summarizer with ElectricSQL, exploring lazy syncing and real-time progress updates.
LLMs finally enable automated testing for documentation by asking questions and asserting on responses.