It’s a trend. More and more tech founders either have their own pet AI projects or are backing someone else’s. Elon Musk has xAI and Grok. Sam Altman, a former president of Y Combinator, cofounded OpenAI. Reid Hoffman is a founding member of Inflection AI. And Jack Dorsey, co-founder of X and founder of Block, has Goose, which he launched in January 2025.
The open-source AI assistant framework, codenamed “goose”, is designed to make building AI agents easier for developers and organizations. The free platform enables users to plug large language models (LLMs) into real-world applications as autonomous agents. In its debut, Goose is demonstrating an ability to handle software development tasks automatically. It’s already generating significant community engagement. With over 10,000 GitHub stars, 709 forks, and 89 contributors within weeks of launch (as of March 17), Goose is emerging as a significant contributor to the open AI ecosystem.
Why does this matter for R&D outside of software development?
Well, the ability of Goose and similar agentic systems dovetails with a trend underway in lab science as well. An October 2024 article in Cell, for instance, describes the potential of “AI scientists” in the domain of biomedical research for targeted problem-solving. The aim is accelerated workflows. Companies like Google are getting in the game with what they have dubbed an AI co-scientist.
The software space is ahead of the curve. According to internal trials at Block, it can save developers up to 20% of their time and sometimes complete “a day or two worth of work all by itself, unsupervised.” This automation includes code migrations, refactoring, testing, and maintenance tasks that traditionally consume significant resources.
While Goose currently focuses on software development, it exemplifies what the Cell authors classify as “Level 1” autonomy—AI assistants that execute specific tasks defined by researchers.
Hacking away tedium
Goose’s initial focus is on assisting software engineers by automating tedious and time-consuming tasks. According to the company, the AI agent can handle jobs like code refactoring and migration, testing, and even multi-step devops tasks on its own. For example, Goose has been used to:
- Convert or refactor code from one framework or language to another (e.g. migrating an app from Ember.js to React, or translating code from Ruby to Kotlin).
- Perform maintenance and optimization tasks such as running performance benchmarks for build processes and increasing test coverage to a specified threshold.
- Generate boilerplate code and tests, scaffolding new API components or creating unit tests for features with minimal human input.
In internal trials, Goose even managed to rewrite entire software platforms in different coding languages, a feat that would ordinarily take human teams days of work. Block’s engineers report that Goose can save developers up to 20% of their time by autonomously handling such laborious tasks. Operating as a true “AI agent,” Goose runs within a development environment to not only suggest code but execute it: reading and writing files, running tests, installing dependencies, and iterating on its own output.
Open-source foundation and features
Block released Goose under the permissive Apache 2.0 license, meaning any individual or company can use, modify, and integrate it at no cost. Jack Dorsey has been a long-standing advocate for open platforms.
From a technical standpoint, Goose is built to be highly modular and model-agnostic. Developers can choose any major LLM – from OpenAI’s GPT-4 to open models – as the intelligence behind their Goose agents. This flexibility allows mixing and matching models for optimal results. Goose also supports a standardized interface for connecting to external tools and services through the Model Context Protocol (MCP), a set of APIs that let AI agents interface with various systems where data and work happen.
In practice, this means a Goose-based agent isn’t confined to one app: it could simultaneously pull data from, say, Google Drive and Slack, then take actions across both. “Goose’s advantage is its ability to work across different systems… It’s not just limited to Google Drive – it can also integrate Google Drive with Slack, for example,” Jackie Brosamer, Block’s VP of Data & AI Platform Engineering, told VentureBeat. Brosamer described how a busy staffer used Goose to summarize a week’s worth of meetings spread across multiple apps.
While Goose’s first mission is to assist software engineers, Jack Dorsey’s team envisions a much broader impact. The company is already exploring non-engineering applications for Goose agents, from streamlining creative processes like music composition to enhancing personalized e-commerce experiences. In essence, any repetitive or structured process that an AI agent could learn to navigate is a candidate for Goose. Block has made it clear it doesn’t intend to monetize Goose directly; the project is fully open-source.
