Guides
Where the catalogue pages answer one question each, these argue something. Every one is built from material we have: what developers said in public about their own projects, an incident in our own repository, or a measurement we ran.
what an article here has to clear
- Built from something real — a quote, an incident, a survey, a measurement — never a summary of the other posts on the query.
- Says what works without us, and how far it gets you, before it says what we do.
- Names what we have not measured, instead of implying we measured it.
Spec-driven development, and what happened to it
The definition, the tools, the people who tried it and stopped, and the problem that was still there when they stopped.
Spec-driven development with Claude Code
What Claude Code already gives you for working from a spec, how far each piece gets you, and how to hand it a spec it cannot rewrite.
The vibe-coding wall, in five people's words
The wall is not the prototype. It is the second month, and it is five distinct failures that get lumped together as one.
Your agent will "fix" the thing you did on purpose
An incident in our own repository: the agent did its job correctly and broke something anyway, because the reason was in a person's head.
Nobody reviews a 2,000-line AI diff
Generation got two orders of magnitude faster. Comprehension did not. That asymmetry is the whole problem, and reviewing harder does not close it.
Your agent is not a yes-man — your prompt is missing a spec
Agreement and invention are different failures, and the advice that fixes the first one does nothing about the second.
A PRD template your coding agent can actually follow
The template, a filled example, and the five places a PRD written for people breaks when the reader is an agent.
What is coming
The next batch is measured rather than argued: three spec-driven tools run against the same feature, and how far a CLAUDE.md drifts over twenty sessions. They publish when the measurements land, not before.