Evergreen: Your PR Gardener
Evergreen helps keep pull requests green and mergeable by fixing CI failures and merge conflicts, and keeping PR branches up to date with the base branch.
GitHub Next
A team of researchers and engineers at GitHub prototyping the tools, workflows, and ideas that will shape the future of software development.
Live Last read · last published · next check
Latest posts
Can agents be proud of their work?
Can agents be proud of their work?
Agentic Workflows Can Use Local Inference
Agentic Workflows gives you complete control to customize your Actions runtime and choose your AI model. You can deploy Agentic Workflows to a custom runner of your choosing and use any compatible AI host or model-routing platform. You can even use local inference running on your Mac!
/goooooooal!
Turn an issue into an agentic mission
Crane: verified code migration
Crane is a migration assistant for GitHub repositories that plans, executes, and verifies code migrations in small agentic steps while keeping humans in control.
Agentics Beyond Code
What happens when you give PMs, compliance teams, and leaders their own agents? A tour of Agentics Beyond Code — an open-source set of GitHub Agentic Workflows for the non-engineering roles that ship, govern, and operate products.
Control what your agentic workflows see with integrity filtering
GitHub Agentic Workflows filter untrusted GitHub content before it reaches the agent. Here’s why integrity filtering matters for repository maintainers, and how we built it.
Animating the native traffic lights in Ace
Animating the native traffic lights in Ace
Agent Functions
Prompts are programs. You wouldn’t write a complex program completely from scratch, in a big, soupy loop without subroutines, and then write it again the next time you wanted to run it. Why let your agent work that way?
A loader with level of detail
A loader with level of detail
Loading states that bleed outside the window
Loading states that bleed outside the window
Canary: a harm gate for agentic systems
Canary puts a small, auditable gate in front of agentic workflows so untrusted artifacts are classified before powerful agents act on them.
The Impact of Automated Repository Maintenance Assistance
The Impact of Automated Repository Maintenance Assistance
Agents are power tools
A practical mental model for agents, workflows, and human-machine systems in agentic engineering.
Agency is the New Resilience
Agents can power robust workflows by intelligently reacting to unexpected conditions, creating a new form of flexible resilience.
Understanding Repositories as Human/Agent Knowledge Factories
Understanding Repositories as Human/Agent Knowledge Factories
Autoloop is porting pandas to TypeScript
tsb is a from-scratch TypeScript port of pandas, being built almost entirely by Autoloop — one iterative improvement at a time.
Repo Mind Light
Holistic repository understanding for humans and agents.
New site, who dis?
New site, who dis?
Autoloop
Automate research, development and anything else with a simple, goal-driven loop
Lean Squad: Exploring Automated Software Verification with Near-Zero Human Labour
Lean Squad: Exploring Automated Software Verification with Near-Zero Human Labour
One Developer, Two Dozen Agents, Zero Alignment
Why we need collaborative AI engineering tools, and a tour of Ace — a research prototype from GitHub Next that brings teammates and coding agents into one shared, multiplayer workspace.
Repo Mind
Global codebase understanding for humans and AI agents.
Start Your Day With Code That’s Better
Start Your Day With Code That’s Better
Adding Weighted Task Selection to a GitHub Agentic Workflow
Adding Weighted Task Selection to a GitHub Agentic Workflow
Repo Assist: Crunching the Technical Debt with GitHub Agentic Workflows
Repo Assist: Crunching the Technical Debt with GitHub Agentic Workflows
Automate repository tasks with GitHub Agentic Workflows
Automate repository tasks with GitHub Agentic Workflows
Generative AI and Changing Inputs
Generative AI and Changing Inputs
Towards Semi-automatic Agentic Performance Engineering
Towards Semi-automatic Agentic Performance Engineering
Intent, meet Toolchain
Intent, meet Toolchain
What Kind of Programming is Natural Language Programming?
What Kind of Programming is Natural Language Programming?
On Continuous AI for Test Improvement
On Continuous AI for Test Improvement
On Natural Language Programming
On Natural Language Programming
Agentic Workflows
Towards Natural‑Language Programming for GitHub Actions
Project Copernicus
Exploring LLM-powered navigation for your codebase
Continuous AI
Exploring LLM-powered automation in platform-based software collaboration
Introducing “Continuous AI”
Introducing “Continuous AI”
Discovery Agent
Agentic Setup, Build, and Testing of Repositories.
Extract, Edit, Apply
An exploration of a new category of assists for using natural language in software development.
Mosaic
Can we derive personalized design systems from sources of inspiration?
Learning Sandbox
Can we make it easy and fun to learn as we build? We’re exploring ways to create personalized, interactive learning environments that integrate into your daily workflow.
GitHub Spark
Can we enable anyone to create or adapt software for themselves, using AI and a fully-managed runtime?
Vitale
Live notebooks in VS Code for JavaScript/TypeScript, web development, and AI experimentation
Copilot Workspace
An agentic dev environment, designed for everyday tasks.
Monaspace
An innovative superfamily of fonts for code. How can we make code more expressive in any editor?
Copilot Next Edit Suggestions
Can we improve Copilot code completion by suggesting the next logical change, wherever it is in your project?
Realtime GitHub
Multiplayer collaboration for your whole repo.
SpecLang
Can we develop software entirely in natural language, and let an AI-powered toolchain manage the implementation?
Bringing the power of AI into your application
A practical introduction to integrating large language models into applications, covering prompt crafting, chaining AI queries, and UX considerations for building AI-powered tools. Also addresses how to measure the impact of AI changes and evaluate whether AI integration is right for a given application.
Code Atlas
How can we make LLM responses more robust and easier to understand by combining their fluid reasoning with rigid structure?