After several successful feature layers, one session lost track of the actual codebase state and started patching against memory instead of files. This post explains the warning signs and why starting fresh was the right

A sidebar property switcher and a background job runner both worked at the feature level. The trouble showed up in everything around them: branch workflow, Playwright selectors, and test environment setup.

This article looks at a session where the AI started by proposing the logging design before touching code. That approach worked well, but two misses still made it through: a redaction rule that didn’t cover the main case

The first version of the article tool handled metadata, content generation, and WordPress publishing across independent rows. Under the surface, it also baked in duplicate-post risks, weak state handling, and product gap

One session was spent comparing docs to the actual codebase, and the mismatch was bigger than expected. The code had moved forward, but the files meant to guide future AI sessions had quietly fallen behind.

This piece traces several production-breaking bugs back to one root cause: the AI knew the APIs, but not the versions actually installed. The fix wasn’t another prompt. It was better build gates.

This article covers an invite system that mostly came together in one session, including roles, expiring tokens, and admin gating. It also shows the two places the implementation fell apart: a subtle Next.js hydration is

A look at where AI helped most in a real SaaS build, from auth flows to security checks and end-to-end tests. It also shows the obvious UI and browser issues it missed until a human called them out.

A build log from running Claude Code and ChatGPT against the same codebase on different machines. The hard parts weren’t code generation, but missing shared memory, stale library knowledge, and system-level setup bugs AI

A real build log from an AI-assisted SaaS project where outdated framework assumptions caused auth, routing, Prisma, and Tailwind problems. It shows what broke, why it broke, and how shared project context reduced repeat