About

I'm Jody Vining. I build software, mostly involving AI. This site is where I write about it.

Right now I'm focused on local AI for coding: running open-weight models on Apple Silicon, benchmarking them against cloud APIs, and figuring out what actually works for autonomous, agentic development. Not autocomplete. Full plan-to-execute coding with local models.

Background

My day-to-day is full-stack development: web apps, iOS apps, databases, APIs, deployment pipelines. A few years ago I started integrating large language models into everything I built.

For an IP law firm, I built a platform running 48 coordinated AI agents that handle content creation, SEO monitoring, competitive intelligence, and compliance tracking. It runs autonomously with a full admin dashboard for oversight.

For a photography business, I built a 33-agent content system that generates blog posts, manages SEO, and handles social media scheduling.

On the side, I've built Thrift Hunter (an iOS app using computer vision for resellers), TopoMeshLab (turning map polygons into 3D-printable terrain), and a few other things I'll probably write about here eventually.

Why This Site

Cloud AI coding tools are getting more restricted: rate limits, subscription changes, pricing adjustments. I started looking seriously at running models locally, and realized the specific niche of agentic coding with local models is barely covered. Lots of people talking about autocomplete and chat. Almost nobody documenting what it takes to get a local model to autonomously write, test, and iterate on real code.

So that's what I'm writing about. Real benchmarks, real costs, honest results. The stuff I'd want to read if I were making the same decisions.