Welcome to GitHub Copilot Day: the future of agentic engineering

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Welcome to GitHub Copilot Day: the future of agentic engineering · Channel: GitHub · 04:27 · Watch on YouTube

Topics: AI agents, Software engineering with AI, AI developer tools, Model inference and optimization

Summary

Kyle Daigle opens GitHub Copilot Day by introducing new Copilot capabilities and inviting both existing and returning users to explore them. He describes how the growth of code and the rise of agents are changing software development, while emphasizing the continued importance of quality, security, trust, and cost. He presents Hydro Fusion, a research preview that selects how models work on a task, and shares its reported evaluation results. He also outlines the shared runtime behind Copilot tools and previews demonstrations of agent routing, reusable skills, repository coordination, and the Copilot SDK.

Key points

GitHub Copilot Day was created to showcase what GitHub has been building before GitHub Universe.
Monthly commits on GitHub more than doubled since April, from 1.4 billion to 2.9 billion.
As agents take on more of the development loop, developers shift from producing every line to shaping what gets built.
Quality, security, trust, and cost remain important as agent use grows.
Hydro Fusion is a research preview designed to choose how models are applied to a task.
Hydro Fusion can use one model, escalate from an efficient model, or have one model draft while another critiques and improves the result.
In offline Terminal Bench 2.1 evaluations, Hydro Fusion improved quality by 4.9 points and reduced estimated cost by 67% compared with Opus 5.
A shared runtime powers the Copilot CLI and GitHub Copilot app, and the Copilot SDK provides a foundation for tools and workflows developers build.
The event will demonstrate model routing, reusable agent skills, repository coordination, and building with the Copilot SDK.

Chapters

Start	Chapter	Summary
00:04	Welcome to GitHub Copilot Day	Kyle Daigle introduces the event, welcomes Copilot users, and describes the rapid growth in GitHub commits.
01:11	How agentic development is changing	Daigle explains how agents shift developers toward shaping outcomes and why quality, security, trust, and cost matter.
02:16	Introducing Hydro Fusion	He describes Hydro Fusion’s model-selection approach and shares its results from offline Terminal Bench 2.1 evaluations.
03:25	The Copilot runtime and event previews	Daigle outlines the shared runtime and Copilot SDK, then previews demonstrations and live coding.

Quotes

Welcome to our first-ever GitHub Copilot day. — 00:04

A prompt becomes parallel work. — 01:11

A new approach designed to take on those decisions for you. — 02:16

And while agents can do more of the work, you still decide what ships. — 03:25

So let’s go build. — 04:27

Concepts and entities

GitHub, Kyle Daigle, GitHub Copilot, Hydrofusion, Terminal-Bench 2.1, Opus 5, Copilot CLI, GitHub Copilot app, Copilot SDK

Full transcript (original language)

00:04 Everyone, I’m Kyle Daigle, the COO of GitHub and the CMO of developer from Microsoft. Welcome to our first-ever GitHub Copilot day. We created this event because, honestly, we couldn’t wait until GitHub Universe next month to show you all that we’ve been building. And while we don’t just want to announce what’s new and shiny, today we’re going to put it to work and show you what’s possible and hopefully give you a few ideas to try yourself. So, if Copilot is already a part of your daily workflow, thank you for building with us. We think

00:38 you’ll see a few surprises today. And if you’re new to Copilot, or honestly, if you just haven’t used it in a while, this is a great day to take another look and join us with the best version of GitHub Copilot yet. If you’re at this event, you already know software development is moving and changing incredibly fast. Since April, monthly commits on GitHub have more than doubled from 1.4 billion commits to 2.9 billion commits a month. That’s a lot more code.

01:11 But, more code alone isn’t the goal. As agents take on more of the loop, the way we build shifts dramatically. A prompt becomes parallel work. A single agent becomes a coordinated system, and our time shifts from producing every line to shaping what gets built. That makes quality, security, trust, and cost more important, not less. Meeting that bar takes more than a powerful model. It takes the right system around it. Context, tools, checks, and a smart way

01:46 to put every single model to work. And that last part is much harder than it sounds. A new model launches every week just about, and they become a part of developer culture. We try them, we compare notes, and we debate our favorites. I do it, too. I did it all weekend. Then, the practical questions arise. Which model should I use for this? For which task? At what cost? And our time isn’t unlimited, and neither are our tokens. That’s why this past

02:16 week we announced Hydro Fusion in research preview. A new approach designed to take on those decisions for you. The idea is simple, even if the engineering around it isn’t. Hydro Fusion looks at the task and decides whether to use one model directly, start with an efficient model and escalate if needed, or have one model draft while another critiques and improves the result. In offline evals with Terminal Bench 2.1, Hydro Fusion improved quality by 4.9 points while reducing estimated cost

02:51 by 67% when compared to Opus 5. You’ve heard us talk a lot about our commitment to model choice over the years, and Hydro Fusion is the next evolution of that. Better quality, lower cost, and a lot less model shopping for you to do. Now, it’s a research preview, so we’re still learning, but the direction is clear. Better results shouldn’t require more complexity or cost more for developers. Now, model orchestration is just one piece of the system. The shared runtime

03:25 now powers the Copilot CLI and the GitHub Copilot app. The Copilot SDK, that same runtime, becomes a foundation for the tools and workflows that you build. And you can start with the agent infrastructure we use ourselves at GitHub, then shape it around your users and code bases, and of course, your way of working. And while agents can do more of the work, you still decide what ships. Now, I know you came to see what this changes in your workflow. So, let’s

03:56 get right into it. We’ll show Hydra fusion routing real tasks, agent skills making proven approaches reusable, and the GitHub Copilot app coordinating across real repositories. Then, we’ll show how you can build on the same foundation with the Copilot SDK. I also know many of you came to watch Burke and Pierce code live. I did, too. And everything before that will give them more powerful tools to build with and more interesting ways to code

04:27 live with all of you. So, everybody wins. Look, your workflow has changed and GitHub Copilot has changed a lot, too. I’m so excited for you to see what we’ve been building. We’ve got a lot to cover today, so let’s go build.