@zaid_mallik1 That’s a great question and one I haven’t been able to fully answer yet – I have an Evel harness and monitoring set up on our generation and indexing to monitor accuracy and regression (models change all the time, and we’ll likely introduce new ones depending on the task) – there’s a codebase Q&A feature that’s largely driven by document information so I have feedback levers to get user input so we can fix things, but my personal My guess from experience is the biggest issue/difference ranking/prioritization with agentic tooling over code generation is that Moxie Docs is not intended to replace in memory systems that can help there, rather it is meant to enhance and improve any workflow or tooling you already have.
If you ask an agent to add a new API route Moxie Docs gives it the location of the relevant documentation (skip the grappling steps), it prompts it on codebase conventions, and it instructs the agent that if any changes affect those docs – it should update the docs in the same PR it makes the change. In theory (and in practice!) all of this should mean better quality code output, but if one really trusts AI agents completely is doing so they will definitely want some type of memory. / Reference retrieval or similar skills established.
<a href