24 Days, ~1,250 Commits: What the Agents Got Right and Wrong
A honest retro on building this blog with agents in the loop — where they saved days, where they wandered, and the guardrails that made the speed safe to keep.
A honest retro on building this blog with agents in the loop — where they saved days, where they wandered, and the guardrails that made the speed safe to keep.
Static pages that update the moment I publish. How a Sanity webhook triggers on-demand revalidation so a fully static Next.js site never serves stale content.
Every post needs a hero image and I'm not an illustrator. How a script rendered decent heroes from HTML — and why that's a placeholder, not the finish line.
Tailwind v4 moved config into CSS, and that made an agent a far better collaborator. How design tokens in one @theme block kept the whole site consistent.
Portable Text is just JSON until you render it. How I mapped Sanity's block format to React components — headings, links, and real code blocks — in the App Router.
One place owns every query, and it's typed end to end. How a groqd service package turns GROQ strings into functions an agent — and the compiler — can trust.
I didn't hand-write the Sanity schema — I argued about it with an agent until the shapes were right. How a conversation became a typed content model with real constraints.
One markdown file at the repo root turns scattered AI help into a fleet that stays on-rails. How CLAUDE.md encodes the house rules every agent reads before it touches a line.
How an idea over coffee became a production blog in 24 days and ~1,250 commits, built mostly by AI agents — with a map to every post that tells the story.