GitHub

We could tell you about how Copilot CLI can optimize your time, but we'd rather have Cassidy Williams sing about it instead.

If you’ve built a multi-agent workflow, you’ve probably seen it fail in a way that’s hard to explain. 🤔 The system completes, but somewhere along the way, something subtle goes wrong. Why is that? Because the moment multiple AI agents start handling related tasks (like triaging issues, proposing changes, opening PRs, etc.) they start making implicit assumptions about state and ordering. Through our work on agentic experiences at GitHub, we’ve learned a core lesson: Multi-agent systems behave much less like chat interfaces and much more like distributed systems. If you want them to be reliable, you have to treat them like code. Here are the three engineering patterns that make that happen: 1️⃣ Typed schemas: Natural language is messy. Add structure at every boundary so agents pass machine-checkable data. Treat schema violations like contract failures. 2️⃣ Action schemas: Vague intent breaks agents. Define the exact set of allowed actions (e.g., assign, close, escalate) so the outcome always resolves to a specific, explicit choice. 3️⃣ Model Context Protocol (MCP): Schemas define structure, but MCP enforces it. It acts as the contract layer, validating inputs and outputs before execution so agents can't invent fields or drift across interfaces. Design for failure first. Validate every boundary. Constrain actions. Get our full guide for engineers on building reliable, scalable multi-agent systems. 👇 https://lnkd.in/dUPnt7ny

Turn a WhatsApp message into a video. 📱🎬 We connected the Copilot SDK to Remotion to build a tool that generates a high-quality promo video in 5 minutes—triggered right from your phone. All thanks to pluggable, portable code. What will you build with the Copilot SDK? ⬇️ https://lnkd.in/giThByAa

Most developers already live in the terminal. And now, so does Copilot. 💻 We initialize projects, run tests, debug CI failures, and make fast changes in the terminal. GitHub Copilot CLI fits seamlessly into that environment, helping you move from an idea to reviewable diffs without context-switching. But the real magic isn't just writing starter code for you. It's how it fits naturally into the way you actually create and ship software. If you're building with AI, a helpful mental model looks like this: 💻 1. The CLI (For Momentum): Start with intent, not scaffolding. Tell Copilot what you want to build, or ask it why a test is failing right at the point of failure. Use the terminal to explore ideas, iterate fast, and get unstuck with low ceremony. 🛠️ 2. The IDE (For Precision): Eventually, speed matters less than exact architecture. Move into your editor to refine APIs, handle edge cases, and make design decisions you’ll defend in review. 🐙 3. GitHub (For Durability): Once the changes look good, use the CLI to commit and open a PR. Now the work is testable in CI, reviewable by your teammates, and ready to ship. The best part? You stay in control. Copilot CLI proposes plans, structures, and diffs, but it never silently runs commands without your approval. It’s a tool for momentum, not a replacement for judgment. Check out our full guide on building a practical Copilot CLI workflow. It includes a GitHub Skills exercise so you can try it yourself in a preconfigured, safe sandbox. 👇 https://lnkd.in/dE6Y7nNZ

Being a team player means learning to let go. What tips do you have for empowering your team? https://lnkd.in/gqNjhNAe

At GitHub Universe, Andrea talks with Ron Efroni (Flox/NixOS Foundation) and Rob Ferguson (Former Microsoft for Startups) about environment hell, why containers don’t fully solve reproducibility, and how Flox builds on Nix for portable, secure, deterministic environments—plus Series B plans, Kubernetes support, CUDA distribution, and the Microsoft for Startups program.   

John Leider, the creator of Vuetify, joins us to talk about his incredible journey as an open source maintainer and show us how to use Vuetify to build beautiful UIs! Bring your questions and curiosity! Project: https://lnkd.in/exM8y_pN

I'm thrilled to announce the Open Source Assistive Technology Hackathon! Join us on May 21–22 at the GitHub headquarters in San Francisco (SF) to celebrate Global Accessibility Awareness Day. Join us for this free, 2-day, in-person event focused on empowering participants to build skills and make real contributions to the assistive technology (AT) tools people rely on every day. New to open source or GitHub? Join us for the GitHub Learning room to learn about core GitHub contribution workflows and practice navigating repositories, issues, pull requests, and code reviews with confidence. This workshop led by Jeffrey Bishop will also cover NVDA and keyboard-only navigation! Ready to contribute? Join us to contribute to one of our featured open source AT projects! We will have maintainers and collaborators available to help. Featured project: NVDA - you’ll also have a special opportunity to collaborate with members of the NVDA team, who will join remotely to help you get set up, answer questions, and review contributions to NVDA. We’re seeking additional projects. If you maintain (or contribute to) an AT project and would like your project to be part of the hackathon - whether it needs docs improvements, triage support, testing, or small starter issues - please reach out to me. Got a new idea you want to work on? Bring it and work on it with accessibility and technology experts! No need for technical skills! Sign up using the event form: https://lnkd.in/eGcxVRt9 #OpenSource #Accessibility #GAAD

Get those tasks out of your head and organized with GitHub Issues and Projects. Not sure how to get started? Follow along with our step-by-step tutorial. https://lnkd.in/d45eUsiR

🆕 OpenAI GPT-5.4 is now generally available and rolling out in GitHub Copilot. Early testing shows ➡️ It consistently hits high rates of success ➡️ Enhanced logical reasoning and task execution for intricate processes Try it out in Visual Studio Code or Copilot CLI. https://lnkd.in/gPKWHeTj