11 pieces
Research
Landscape scans and working maps. Researched by Claude under my direction; I review everything before it ships. Surfacing what my second brain is building on is part of the point.
- № 011 Jul 30, 2026
There is no audit command, because auditing stopped being a command
Looking for a slash command that audits a whole codebase turns up nothing. The absence is the interesting part - that work moved to fan-out, not a verb.
tooling
4 min - № 010 Jul 30, 2026
The best design engineer portfolios are not documents about work, they are the work
Design engineers proved a discipline by shipping the artifact instead of describing it. What that changes about seniority, evidence, and the case study.
design-engineering
6 min - № 009 Jul 30, 2026
Better is a statistical claim, not a vibe
Most teams judge conversational AI by reading outputs and trusting the feeling. The alternative is error analysis, a calibrated judge, and a powered comparison.
evaluation
8 min - № 008 Jul 30, 2026
The most agent-native knowledge base is a folder of files
Every note-taking tool is racing to ship an MCP server. The ones that stored plain files already had the integration and did not need to build it.
tooling
6 min - № 007 Jul 26, 2026
The documentation skill ecosystem is crowded, and the quality gate is empty
Dozens of repos will teach an agent to write docs. Almost none will stop it shipping bad ones. That asymmetry tells you where the real work is.
tooling
6 min - № 006 Jul 13, 2026
Vercel Eve makes an agent a folder of files, and the AI Gateway is optional
Vercel's Eve is an open-source, filesystem-first TypeScript agent framework. It defaults to the AI Gateway but does not require it.
agentic-systems
8 min - № 005 May 24, 2026
An agent config file is a budget, not a document
CLAUDE.md and DESIGN.md both fail the same way: people write them as documentation. They are instruction budgets, and every line you add makes the others weaker.
tooling
8 min - № 004 May 24, 2026
The platform absorbs the framework, and methodology is what survives
Claude Code shipped natively almost everything Superpowers existed to provide. Watching its author delete his own code in response is the most useful thing in the category.
tooling
7 min - № 003 May 23, 2026
Claude Opus 4.7's house style is already editorial — the work is restraint, not extraction
Which AI design skills actually produce high-taste editorial frontend output, and why the answer is mostly already inside Claude.
ai-tooling
7 min - № 002 May 23, 2026
Divergence is nearly free now — the bottleneck is judgment
How the design community is operationalizing diverge-then-converge workflows with LLMs in 2026, and why convergence is now the only interesting problem.
ai-delivery
9 min - № 001 May 23, 2026
Snowflake Intelligence is the floor — Cortex Agents REST API is where the real work starts
How Snowflake Cortex Agents actually works as a stack, what Snowflake Intelligence restricts on the UI layer, and how to handle Vega-Lite charts in a Next.js generative-UI build.
agentic-systems
9 min