👋 Hey all,
I haven't used this platform often, my last real blog post was about me failing at running a startup - so it's exciting and scary to say I'm at it again!
Over the last few months I've been quietly thinking about how AI is changing the field of software development, and what that means for engineers. I've been building full-stack applications (and desktop, and mobile, and some other stuff) for around 8+ years now, and the changes I've seen in the last year alone are such a fundamental shift in the industry - and how engineers work - that I thought about what kinds of issues I've had adopting new technology.
My Agentic Coding Workflow
I'm going to keep this short and avoid buzzwords, in fact I'll just give a TL;DR list of how I've been using AI:
- PR Code reviews, 100% of the PRs I open are reviewed by an agent alongside humans, I use greptile for this at my work and CoPilot / Gemini (antigravity) on my personal plans for side projects.
- Small front-end, well-scoped features. AI excels at highly specific tasks with well defined context & conventions.
- PR descriptions, my least favorite part of development is writing commit messages, 100% of my messages are authored by AI now.
- Testing. To some degree unit tests are great candidates to offload to LLMs, but also things like storybook stories for UI libraries where they aren't necessarily user-facing and are low risk, I delegate to agents.
Where things don't work well...
The biggest pain point I've found adopting AI tooling in my day to day workflow is the amount of loops / iterations needed to get a "ready to go" final implementation.
I found myself continually typing "follow our codebase conventions" in prompts, pointing it to example files to use as a reference, and then prompting it to update documentation (if I remember).
Enter: Moxie Docs
I developed Moxie Docs after much iteration and frustration. I had dissolved the last startup I attempted (after around 7 years of customers using it!) and it was quite the walk down memory lane recreating Stripe, buying domains, setting up marketing flows, auditing SEO. The million little things that go into trying to build a product for users.
What does it do?
In short - it solved a problem for me (and I imagine many others) by truly creating living documentation in your codebase, automated developer docs sitting next to the code itself. Auto-updated, auto indexed, checked on every PR merge. Alongside that we have an MCP for users using AI agents - connect Moxie Docs to Claude Code, Cursor, Codex, etc. and we give the tools and guidance for AI workflows to detect what changes impact docs, update them automatically in the same PR they are making changes in, and pull in codebase conventions before starting changes.
Every API file requires a test? Components split into domain folders? Re-use templates for back-end routes? File name conventions, organization, all of it is detected & continually updated by Moxie Docs and surfaced to agents.
A short list of features:
- Moxie Docs MCP - agent-ready context continually updated as your codebase changes
- A changelog viewer / copy feature tied directly to PRs, organized by internal and external changes
- We detect orphaned documentation, documentation drift, and missing docs. We give you 1-click buttons to generate a PR from Moxie Docs into your codebase - yours to review and accept.
- PR Description alignment - we can automatically categorize PR types, rewrite & summarize them to keep formats / templates consistent, easier for engineers to know what's going in / what to review.
It's been a lot of fun to build & launch - and I've gotten some great feedback already :) would love to answer any questions or hear thoughts if you'd like to check it out: Moxie Docs
Top comments (0)