Ryan Eberhard

Build Log

I write about how I operate — the systems, the decisions, the failures. If you want to know how I think before we talk, start here.

• Your AI Agent Doesn't Know What to Do Next Feb 2026

  Most agent builders focus on what the agent can do. Nobody talks about what it should do right now. 31 detectors, numeric scoring, and pre-generated recommendations — the difference between an assistant and a colleague.

• I Built an AI Design Critic That Judges My Apps Against Apple's HIG Feb 2026

  An automated design quality system that uses Claude to read a prose design system as config — no hardcoded rules. Code analysis + visual screenshots + a full-app crawler, enforcing Apple-grade standards from commit to deploy.

• Compiled Context: How I Stopped Fighting Agent Memory and Started Engineering It Feb 2026

  A 350-line Python script reads Slack history, Obsidian docs, feedback logs, and a unified inbox — then compiles a single file per agent every 30 minutes. Three agents boot with full domain awareness. No retrieval. No queries. No luck.

• My AI Controls My Webcam Feb 2026

  I asked Claude to fix my webcam image. It captured a frame, analyzed the lighting, adjusted the hardware settings, and iterated until it was balanced. No app, no sliders — just a conversation.

• The CI Pipeline I Actually Want Jan 2026

  Two stages, four surfaces, zero drama. A CI pipeline for iOS, web, backend, and database — informed by Fowler, Beck, and DORA — that makes autonomous AI deployment rational.

• How I Made Claude Code Do Its Own Deployments Jan 2026

  Most developers use AI to write code, then manually run migrations and deploy. Give Claude explicit permission via an SOP document, and the entire backend-to-production loop happens in one conversation.