Spec-Kit versus RPG for Code: How AI Assembles Entire Projects Using Graphs

[deepak-vij]

@mtodd @tpope @jonmagic @mutle @cdb

Hello, I recently started playing around with spec-kit. I must say, very impressive work indeed.

On the similar line, just yesterday I came across the following paper from Microsoft, RPG for Code: How AI Assembles Entire Projects Using Graphs. This looks very similar to spec-kit.

Essentially, overall thought process is similar. (1) Proposal-Level Construction organizes and refines requirements into a functional graph by retrieving nodes from a large-scale feature tree. (2) Implementation-Level Construction expands this graph into the full RPG by encoding file skeletons, interfaces, and data flows. (3) Graph-Guided Code Generation

The key difference being the fact that in spec-kit intermediate plans are externalized (using markdown files) and refined across iterations to connect intent with execution versus RPG proposes a graph based data structure for much better code generation efficiency.

Any thoughts on this and how soon can we transition towards graph based approach planning approach versus based on markdown files. This looks quite promising. Thanks.

https://arxiv.org/pdf/2509.16198