Sep 10, 2026 · 21 min · 10 segments
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…
Iury SouzaGuestNicola CortiHostI've been using OpenCode sometimes, but I mostly use the one that the default model will provide now.
And so, yeah, let's deep dive into what is a harness, why people would want to use it, and...
So I think there's a bunch of definitions around hardnesses, but I think it's basically the things around the model.
So what makes that loop works? Because I think I remember back in the day, like, I think it was Goose from Block.
And they had a blog post and some docs around it, like, explaining how the loop would work, like, tool calling and injecting context and doing the whole thing.
So, when I saw that, I realized that, okay, like, you can customize these things quite a lot.
Like, so just extending on top of what they give you already, which is, like, skills, custom skills, like, how do you handle context injection on the sessions, like, handing off work, compaction, this kind of stuff you can already customize quite a lot.
And then there's like when you go to something like Pi, which offers like extensions and being able to build tools and hand tools directly to the agent, then you get even like it becomes very interesting how you can.
It's basically how you do like the context engineering, but more in a deterministic way, I think.
I've been using OpenCode sometimes, but I mostly use the one that the default model will provide now.
And so, yeah, let's deep dive into what is a harness, why people would want to use it, and...
So I think there's a bunch of definitions around hardnesses, but I think it's basically the things around the model.
So what makes that loop works? Because I think I remember back in the day, like, I think it was Goose from Block.
And they had a blog post and some docs around it, like, explaining how the loop would work, like, tool calling and injecting context and doing the whole thing.
So, when I saw that, I realized that, okay, like, you can customize these things quite a lot.
Like, so just extending on top of what they give you already, which is, like, skills, custom skills, like, how do you handle context injection on the sessions, like, handing off work, compaction, this kind of stuff you can already customize quite a lot.
And then there's like when you go to something like Pi, which offers like extensions and being able to build tools and hand tools directly to the agent, then you get even like it becomes very interesting how you can.
It's basically how you do like the context engineering, but more in a deterministic way, I think.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.
