我们作为开发者的工作正在改变。 我们从写每行代码到与AI助手合作，但目标并没有改变：飞船完成了软件工作。 一个功能，一个错误的修复，一个安全修复。 已完成的工作，而不是更多的代码，才是重要的产出。
今天在GitHub宇宙，我们分享了我们对GitHub作为世界上最全面的软件传输系统的愿景。 我们的推出产品将这个愿景带入生命，并展示我们如何重建其下的基础设施。
我们的愿景：从想法到完成的软件
SDLC不再是从规划到编码到测试到部署的线性路径。 代码的更改，同事的请求，或生产量度可以触发系统中任何地方的工作。 我们大多数(和我们的agents)跳跃在不同阶段之间，这取决于手头的任务。 开发者确定方向，作出决策，并对哪些船只负责。 Agents执行并验证更改。 所有东西都从不断学习的机器中进进出出，使下一个周期更好。
这个代理软件发送系统解决了三件事：
速度：尽可能快速、高效和符合成本效益地从想法到完成软件
智能：将跨平台的背景与连续学习引擎结合，这样系统在使用时会变得更好。
质量：默认保证完成的工作安全，在发船前抓捕错误
GitHub是为开发者及其agents建造的。 我们也在提供基础原始物，你继续控制工作工具和成本
推进工作的新特点和新产品
今天在宇宙中，我们展示了一些最新的产品和功能，从GitHub Copilot应用开始。
GitHub Copilot app：从发行到合并，在一个app中
GitHub Copilot应用从地面上设计，用于代理软件的发送。 它连接了您的 agents 和工具跨项目，这样就可以在不进行每次交接的情况下引导工作。 这不仅仅是生产代码：你可以写谱，介绍设计讨论，或者在一个agent-第一表面分析遥测。
Copilot 融合：最佳多型协奏曲
不同的任务需要不同的模型。 开发者不必管理每个模型选择，Copilot Fusion评价一个任务需要跨越推理，代码生成，调试，以及工具使用，然后选择如何跨一个或多个模型执行。 在人工分析编码Agent指数中，Fusion Max在Opus 5.5 Max的两个点内以成本的一半和每任务减少40%的时间得分。
Copilot项目：给Copilot一个目标
在今天开始的 GitHub Copilot 应用程序中可以本地使用，Copilot 工程允许您在完全管理，并行的会话中运行 agent 工作跨多个重置程序。 你有一个地方可以跟踪进展，做决定，并指导接下来发生的事情。 也就是说，您可以少花时间去管理执行，多花时间来决定什么值得建设。
会话连续性：继续跨表面工作。
当您切换设备或工具时，您的工作不应该重新开始。 会话连续性允许您开始并跟随同一云 agent 会话跨越 GitHub app, VS 代码，CLI，和网络。 从你的手机开始站立，从你桌上的VS代码引导，午餐时从github.com检查。 agent不断在云中工作。 我们的多型机车能够使其强大、快速和具有成本效益。
Copilot自动化：学习结果
Copilot自动化可以让你在Repo内创建AI动力的工作流程。 他们跑在云里，你可以手动踢他们，或者设置他们来响应一个事件或者按时间表跑。 这使得您可以协调agents，处理诸如发行分类和代码审查等任务。 现在，自动化可以学习它们的成果，并提出改进建议，供你们审查和批准。 这是实际学习的引擎。 利用已完成工作的结果，使下次运行更加顺利。
Copilot代码审查 Max：更深入地审视哪些船只
更多的代码意味着更多的拉请求和更多的审查需求。 如果我们能保持舰只的质量才能更快前进 Copilot代码审查浮出有意义，可操作的bug，开发者需要捕捉。 今天，我们推出 Copilot代码审查马克斯，它在Review Bench上贴出了最高的F1分，用于高热度问题。 文学与平衡保持日常评论的快速性；当变化需要时，新的马克思提供了更深入的视角。
没有单独企业订阅的 Copilot 特性
从今天开始，GitHub Entertainment 客户可以使用所有Copilot特性，而无需单独购买订阅。 包括GitHub Copilot应用，代码审查，自动化等等。 你付的人工智能信用。
这个通过今天的GitHub销售可以提供，自服很快就来了。 现有的Copilot Business和Copilot Entertainment 客户可以保留目前的计划，或者在准备就绪时搬过来。 GitHub也保持开放和可扩展。 继续使用你们团队所依赖的工具，包括Cloudflare,Datadog,Figma等。
我们的基础设施投资
所有这一切都取决于一个快速、可靠和可以扩大的基础。 Agents正在改变软件开发的音量和形状。 增加能力很重要，我们也不得不改变结构。
今年至今为止
400万个CPU核心
160 PB 高速存储
50Tb/s网络能力
我们正向Azure的多地区部署，我们期望明年年初完成移民工作。 这些投资为我们提供了更大的增长空间。 下一步是建立处理工作的系统。
因为GitHub的活动已经增加，代理时代要求对吞吐量和可靠性实行不同的上限。 三条原则指导我们新的Git架构：简化原始体，解密存储与计算，尽量减少协调。
Git只是我们推广平台工作的一个例子。 GitHub的人工智能网关
超过前1%
我们的工作正在改变，但开发者比以往任何时候都更重要。 我认为我们至今为止所写的所有软件都不到我们未来五年所写软件的1%。 这不仅仅是创造更多代码的机会。 解决我们以前无法解决的问题。
因此，我们正在连接整个软件传输系统，让开发者有更好的方法与agents合作，并为接下来的发展发展基础设施。 而我们是在GitHub上做的，我们可以一起建设。
Jay Parikh是世界上增长最快、最成功的消费者、开发商和企业技术公司的资深执行官、董事会成员和顾问。 作为位于Microsoft的CoreAI执行副总裁，Parikh领导组织建设企业号所需的端对端代理系统，进行船运，运营AI agents. 他的团队，包括GitHub，也建造工具和平台开发人员需要创建下一代软件。
此前，Parikh是云安全启动公司的首席执行官，在此之前，他是Facebook的全球工程和基础设施负责人。 十多年来，他领导了建设和运营支持公司从3亿人增长到30多亿的有形基础设施和软件的团队。 从数据中心、服务器和海底电缆到软件、数据、AI和安全系统，通过在开放计算项目和Telecom Infra项目中的领导作用，他帮助将基础设施创新扩展至单个公司之外，使整个行业受益。
职业生涯早些时候，帕里赫曾在阿卡迈和宁担任领导职务。 带来是一个系统思考者对人类的深刻挑战的看法：如何建立团队，不断学习、移动，并随着世界的变化而实现。
AI、agents、语音和物理控制正在将软件开发从操作工具转向表达意图
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Our jobs as developers are changing. In a couple of years, we’ve gone from writing every line of code to working with AI assistants to orchestrating fleets of agents. But the goal hasn’t changed: ship completed software work. A feature, a bug fix, a security fix. Completed work, not more code, is the output that matters.
Today at GitHub Universe, we shared our vision for GitHub as the most comprehensive software delivery system in the world. We introduced products that bring that vision to life and showed how we’re rebuilding the infrastructure underneath it.
Our vision: From idea to completed software
The SDLC is no longer a linear path from planning to coding to testing to deployment. A code change, a request from a colleague, or a production metric can trigger work anywhere in the system. Most of us (and our agents) jump between different stages, depending on the task at hand. Developers set the direction, make decisions, and stay accountable for what ships. Agents implement and verify changes. And everything feeds into and out of a continuous learning machine that makes the next cycle better.
This agentic software delivery system solves for three things:
Velocity: get from idea to completed software as quickly, efficiently, and cost-effectively as possible
Intelligence: combine context across the platform with a continuous learning engine, so the system gets better as you use it
Quality: make completed work secure by default and catch bugs before they ship
GitHub is built for developers and their agents. We’re also making the underlying primitives available, so you can build your own bespoke solutions. You stay in control of the work, the tools, and the cost.
New features and products that move the work forward
Today at Universe, we showed some of the latest products and features in action, starting with the GitHub Copilot app.
GitHub Copilot app: From issue to merge, in one app
The GitHub Copilot app is designed from the ground up for agentic software delivery. It connects your agents and tools across projects, so you can direct the work without making every handoff. And it’s for more than producing code: you can write specs, present design discussions, or analyze telemetry in one agent-first surface.
Copilot Fusion: Multi-model orchestration at its finest
Different tasks need different models. Developers shouldn’t have to manage every model choice to get the right result at the right price. Copilot Fusion evaluates what a task needs across reasoning, code generation, debugging, and tool use, then chooses how to execute across one or more models. In the Artificial Analysis Coding Agent Index, Fusion Max scored within two points of Opus 5.5 Max at half the cost and 40% less time per task.
Copilot projects: Give Copilot a goal
Available natively in the GitHub Copilot app starting today, Copilot projects let you run agent work across multiple repos in fully managed, parallel sessions. You have one place to track progress, make decisions, and direct what happens next. That means you can spend less time managing execution and more time deciding what’s worth building.
Session continuity: Keep working across surfaces
Your work shouldn’t restart when you switch devices or tools. Session continuity lets you start and follow the same cloud agent session across the GitHub app, VS Code, the CLI, and the web. Start from your phone at standup, steer it from VS Code at your desk, and check in on it from github.com at lunch. The agent keeps working in the cloud. Our multi-model harness that enables it is powerful, fast, and cost-efficient.
Copilot automations: Learning from the results
Copilot automations let you create AI-powered workflows within your repo. They run in the cloud, and you can kick them off manually or set them up to respond to an event or run on a schedule. That lets you coordinate agents to handle tasks such as issue triage and code review. And now, automations can learn from their outcomes and propose improvements for you to review and approve. That’s the learning engine in practice. Use the results of the completed work to make the next run better.
Copilot code review Max: A deeper look at what ships
More code means more pull requests and more demand for review. Moving faster only helps if we can maintain the quality of what ships. Copilot code review surfaces meaningful, actionable bugs that developers need to catch. Today, we introduced Copilot code review Max, which posted the highest F1 score on ReviewBench for high-severity issues. Lite and Balanced keep everyday reviews fast; the new Max provides a deeper look when a change needs it.
Copilot features without a separate enterprise subscription
Starting today, GitHub Enterprise customers can use all Copilot features without buying a separate subscription. That includes GitHub Copilot app, code review, automations, and more. You pay for the AI credits you use.
This is available through GitHub Sales today, with self-serve coming soon. Existing Copilot Business and Copilot Enterprise customers can keep their current plans or move over when they’re ready. GitHub also remains open and extensible. Keep using the tools your team depends on, including Cloudflare, Datadog, Figma, and more.
Our continued infrastructure investments
All of this depends on a foundation that’s fast, reliable, and built to scale. Agents are changing the volume and shape of software development. Adding capacity matters, but capacity alone won’t get us there. We’ve also had to change the architecture.
So far this year, we’ve added:
4 million CPU cores
160 PB of high-speed storage
50 Tb/s network capacity
We’re also migrating to a multi-region deployment in Azure, while continuing to serve incredible levels of traffic from developers around the world. We expect to complete the migration early next year. Those investments give us more room to grow. The next step is building the systems that handle the work.
We’ve scaled Git for years as activity on GitHub has grown. The agentic era calls for a different ceiling on throughput and reliability. Three principles guide our new Git architecture: simplify the primitives, decouple storage from compute, and minimize coordination.
Git’s just one example of the work we’re doing to scale the platform. We’ve done a lot of work across GitHub’s AI gateway, the harness, and much more.
Beyond the first 1%
Our jobs are changing, but developers are more important than ever. I think all the software we’ve written to date will amount to less than 1% of the software we’ll write in the next five years. The opportunity isn’t just to produce more code. It’s to solve problems we couldn’t solve before.
That’s why we’re connecting the full software delivery system, giving developers better ways to work with agents, and evolving the infrastructure for what comes next. And we’re doing it on GitHub, where we can build together.
Jay Parikh is a seasoned executive, board member, and advisor to some of the world’s fastest growing and most successful consumer, developer, and enterprise technology companies. As Executive Vice President of CoreAI at Microsoft, Parikh leads the organization building the end-to-end agentic system required to build, ship, and run AI agents in the Enterprise. His teams, including GitHub, also build the tools and platforms developers need to create the next generation of software.
Previously, Parikh was CEO of a cloud security startup and, before that, was Facebook’s global head of engineering and infrastructure. For more than a decade, he led the teams that built and operated the physical infrastructure and software that supported the company’s growth from 300 million people to more than 3 billion. His teams worked across the stack, from data centers, servers, and subsea cables to software, data, AI, and security systems. Through leadership roles in the Open Compute Project and Telecom Infra Project, he helped extend infrastructure innovation beyond a single company to benefit entire industries.
Earlier in his career, Parikh held leadership roles at Akamai and Ning. He brings a systems thinker’s perspective to a deeply human challenge: how to build teams that keep learning, moving, and delivering as the world changes around them.
AI, agents, voice, and physical controls are shifting software development from operating tools to expressing intent—and helping developers stay focused on building.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.