Development with intent

I have now run one project through Anthropic's intent.md pattern: the RAG app I built in October (cited-rag-agent on GitHub) started from an intent.md, then a spec and a plan, and all three sit in the repo next to the code. The next project I start, mine or a client's, begins the same way.

The idea is out of their SDLC playbook (published 21 August, six stages from Plan through Maintain). The argument is that code stopped being the bottleneck and the human-speed steps around it became the constraint. So stage one is the idea written down as a version-controlled intent.md, and the final stage closes the loop by feeding monitoring output back in as another one.

I would call that development with intent rather than planning, which suggests a document nobody opens after week two. Intent means the reason for the change sits next to the code, so whoever picks it up next, person or agent, reads why before they read what.

What appeals to me is that it's both less rigid and more structured. Less rigid, because there is no committee and no template to satisfy. More structured, because the part that usually stays in my head has to exist as a file before any code does, and it's version controlled, so I can go back later and see what I thought I was doing.

The playbook supplies the measure as well: time from idea to committed intent.md, and what share of intents are accepted into Design. If an idea cannot survive being written down inside an hour, it isn't ready to build.

Source: Anthropic, The AI-native SDLC Playbook