goose

Goose is an open-source agent that lets language models use tools and take action. Block launched it publicly in January 2025.

I led goose’s product experience from its earliest builds through open-source launch and donation to the Agentic AI Foundation. I developed interaction patterns for agentic software, rebuilt the frontend, and aligned the product experience with its launch story.

We had two and a half weeks to make Goose’s desktop app ready for public launch. The app worked, but it looked and behaved like what it was: an engineering-led, volunteer-built frontend with a lot of AI-generated code.

There was no time for a separate design pass. I worked in the code with engineering, added Tailwind so we could move faster, then rebuilt the surface around clearer spacing, type, and component states.

The first-use problem was not chat. It was configuration. I reorganized settings around the controls people needed first, with clearer labels and grouping so setup did not depend on already knowing Goose’s internals.

I added onboarding for provider selection, turning model setup from a settings hunt into the first-run path. Goose launched on January 28, 2025 and reached the top of GitHub Trending.

The launch changed the interface problem. As contributors added features, the original navigation stopped scaling. I moved the app from a narrow chat layout to a wider workspace with sidebar navigation, session history, and search.

Users also needed sessions to belong to projects. I built the project model through the interface and Rust backend so context could be organized above the single chat.

Not every plausible idea survived contact with use. I tried left-aligning user messages and adding message threads because I expected the terminal-like structure to feel familiar. Informal testing proved the opposite. The alignment made people hesitate, so I dropped it.

Explore Goose on GitHub