Field notes from the work: AI coding agents, the Model Context Protocol, SaaS architecture, and what 25+ years of building production software has taught me about doing it well with AI in the loop.
Some AI agents "fix" a failing test by editing the test until it passes, quietly destroying the one signal you were counting on. Here's the setup that keeps them honest, and what changed with Claude Opus 4.8.
READ ARTICLE →The Model Context Protocol is how AI agents call tools, but almost no existing API speaks it. API2MCP wraps any REST API and presents it to AI as an MCP server, no rebuild required.
READ ARTICLE →Vibe coders ship at a pace traditional cycles can't touch, right up until the app needs auth, payments, and AI infrastructure. The answer isn't to compete. It's to embrace and extend.
READ ARTICLE →Lean into Git, keep your schema in sync, move business logic into a rules engine, and break up large files. Four principles that make AI coding agents far more productive, and far less dangerous.
READ ARTICLE →Every app needs the same foundation: user management, payments, subscriptions, AI integration. Why I stopped rebuilding it every time and started Wildwood, a backend-as-a-service platform.
READ ARTICLE →Let's talk about what you're trying to ship and whether we're the right partner.