Jack Dorsey’s Block introduced an artificial intelligence (AI) agent dubbed Codename Goose on Tuesday. The parent company of mobile payments system Cash App, point-of-sale service Square, and music streaming platform Tidal, has now released an open-source AI agent that can handle software development tasks. The agent comes with compatibility for a wide range of large language models (LLMs) and can connect with a large number of applications and tools using extensions. While it can primarily handle engineering tasks for now, the company is exploring non-engineering use cases for the tool.
Table of Contents
Block Releases Codename Goose
In a blog post, the Dorsey-led company announced Codename Goose. It is an AI agent framework, as it can also be used to build new agents that can perform specialised tasks. Since it is open-source, it can be run locally, eliminating data privacy concerns that generally surround AI agents.
Due to its open-source framework, it also provides versatility to developers. Goose can be powered by any LLM, be it open-source models such as DeepSeek-R1 or proprietary ones such as Gemini, Claude, or GPT. This way, developers get the choice to use the AI model they prefer.
Additionally, the AI agent also supports a wide range of extensions. These extensions allow Goose to connect with applications and tools such as GitHub, Google Drive, JetBrains IDEs, and more. The company stated that some of these extensions are already added to the agent’s directory, while developers can also integrate and build newer ones. Notably, all Goose extensions are built on the Model Context Protocol (MCP).
Block also highlighted that Goose can be run locally as a desktop app or through the command-line interface (CLI) using the same configuration.
Coming to use cases, the AI agent primarily handles coding and software development tasks and can handle complex commands. Some of the applications of Goose include code migration (Ember to Reach, Ruby to Kotlin, etc.), transitioning a code base from field-based injection to constructor-based injection, conducting performance benchmarks, creating Datadog monitors, removing or adding feature flags, and more.
Codename Goose can be accessed from Block’s GitHub listing. It is available with an Apache 2.0 licence, which allows both academic and commercial usage.
