Carson Gross’s September 21, 2026 essay proposes a new convention for agentic coding: store selected Markdown in a `/src/md` directory beside the code it helps produce. His central claim is not that prose literally replaces implementation, but that Markdown increasingly carries the intent that ephemeral prompts currently hide.
The proposal treats Markdown as source-like because it is plain text, diffable, searchable, reviewable in pull requests, and natively handled by both people and language models. Gross suggests locating module overviews, feature descriptions, data-model notes, API notes, and infrastructure notes near the relevant code, while leaving higher-level process documentation in systems such as issue trackers or wikis.
He assigns tests a supporting role. Tests should provide automated confirmation based on the Markdown, while the prose remains the human- and agent-readable account of what the system does, why it does it, and what it must not do. That division is intended to make intent easier to review without forcing every design decision into low-level test syntax.
The article is an argument and a proposed convention, not evidence that `/src/md` is a proven engineering standard. Gross says he has not used the structure extensively and expects his thinking to change. He also warns that Markdown needs a “complexity budget,” careful curation, and occasional synchronization with generated code, including changes made back to the implementation.
The practical question is therefore not whether teams should dump every prompt into a repository. It is whether a small, maintained layer of source-level intent can reduce ambiguity without becoming stale clutter. The essay offers a useful answer: keep the Markdown close to the code, make it reviewable, and treat it as a living specification-like artifact rather than a permanent transcript of prompting.