How much of your coding agent can you make your own? Today we’re recording in person during the GDE Summit, with Iury Souza, a mobile developer and platform engineer who loves tinkering with his dev tools.
Iury takes us through his agentic development setup with OpenCode and Pi, from choosing models for planning and implementation to teaching an agent to build its own extensions. We explore what a harness actually does, how skills and AGENTS.md shape the context a model sees, and why sharing your perfect setup can be just as tricky as sharing your dotfiles.
We also talk about the future of terminal interfaces, the reality of running local models on a laptop, and how to keep up when agents produce more work than you can comfortably review. Iury shares his experiments with annotated plans and his open-source Visual Artifact Renderer, which turns agent output into visual explanations, code reviews, and dashboards.
Enjoy the show 👨🍳
Show Notes
- 00.00 Intro
- 00.46 Episode Start
- 01.09 Yuri’s Introduction
- 02.00 What is your AI stack?
- 04.05 Building your own harness
- 05.57 The first thing you change in your harness
- 07.45 Sharing your harness customization
- 10.29 Using Graphical IDEs with AI?
- 12.32 Local models
- 15.46 What’s next for agentic development?
- 19.00 Yuri’s favorite creators
- 20.25 Where people can find you online?
Resources
- OpenCode - Iury’s main coding harness
- Pi - the extensible coding agent behind his experiments
- Visual Artifact Renderer on GitHub - Iury’s tool for visual explanations, reviews, and dashboards
- My 26-Q1 Agentic Coding Workflow - more background on Iury’s setup
- @iurysza on GitHub
- @iurysza on x
- Iury’s Website