Runtime and verification belong in the architecture
AI workflows need explicit verification at every layer.
Getting the first `builds` article live clarified something important for me: AI-friendly website workflows need a hard separation between code state, CMS state, and live production state. A few things that looked like code bugs were really missing or corrupted live records. That means agents need explicit checks at each layer instead of assuming a successful deploy or publish means everything is healthy.
The other durable lesson is that runtime and verification are part of the architecture. Tooling context can mutate a repo before work even starts, and a green build is not an end-to-end signal. If I want agents to ship reliably with EmDash, the workflow needs isolated runtimes, reproducible live-state inspection, and explicit verification gates from schema all the way to browser output.