Jack Dorsey Backs Open Source After Building Bitchat with Block’s AI Agent Goose

Jack Dorsey is lending his public endorsement to open-source AI development. He recently confirmed that he used Goose, Block’s AI agent for developers, to build Bitchat. The app is a personal messaging experiment based on Bluetooth mesh networks. By using an open-source tool to rapidly prototype Bitchat over a weekend, Dorsey underscored the practical appeal of local-first, developer-empowering tooling.

Goose was initially created for internal use within Block. It automates common engineering tasks like project scaffolding, debugging, and test generation. It is designed to run locally or via CLI. The tool connects to any LLM through a modular protocol. Block has released Goose under the Apache 2.0 license. Its codebase is now available on GitHub and open for external contributions.

Jack also called out Qwen3-Coder, an open-source coding model released by Alibaba, in a recent X post: “Goose + qwen3‑coder = wow.”

Qwen3-Coder is a brand new (we covered it yesterday) specialized LLM built for software development and released under the Apache 2.0 license. It has gained attention for its strong performance in agentic workflows and its compatibility with tools like Goose.

Dorsey’s remarks and his hands-on use of Goose reinforce his wider advocacy for open-source AI tools. Goose and Qwen3‑Coder both reflect a shift toward transparent, community-driven agentic tools. These tools are built to empower developers without vendor lock-in.

When Jack Dorsey casually name-drops your AI stack, it's not just a signal. It's a seismic ping in the open-source radar.

Yesterday, I published a Pure Neo Signal on Alibaba’s Qwen3-Coder—a 480B parameter open-source coding model built for autonomous dev agents. My thesis? Open source isn’t just fast. It’s outpacing the incumbents in speed, modularity, and real-world usability.

Then Jack Dorsey did what Jack Dorsey does: dropped a two-part tweet that read like a minimalist manifesto for open AI tooling.

First: “goose + qwen3-coder = wow”
Then the follow-up, almost too clean to be real: “open source moves faster and better.”

Jack isn’t just retweeting vibes. He’s building with this stack. His team at Block made Goose, a local-first agent framework now open-sourced under Apache 2.0. He used it to prototype a Bluetooth mesh chat app, Bitchat, in a weekend. And now he’s shouting out Qwen3-Coder, a model I spotlighted less than 24 hours earlier.

This isn’t coincidence. It’s convergence.

What’s happening here is a new pattern for AI dev:

• Models like Qwen3-Coder offer deep, specialized capability.
• Frameworks like Goose make it pluggable and useful.
• You can now run these models for free on your Mac, with no API key, no tracking, and no cloud bill.
• And founders like Jack are rolling up their sleeves and actually shipping with it.

The cost gap isn’t theoretical. While open-source agents run locally for free, enterprise stacks are veering into absurdity. Salesforce’s AgentForce reportedly clocks in at $2 per chat interaction. Multiply that across a support team or sales funnel, and you’re looking at margins eaten alive by usage fees. Open-source flips that equation. Run Qwen3, Mistral and Kimi2 on proper Macs, wire it up to Supabase and n8n or Pipedream, and you’ve got a AI agents and RAG costing exactly zero. That’s not just cheaper, it’s sustainable, scalable, and yours to own.

Forget glossy demos. This is the gritty, GitHub-driven future of AI tooling.