SEQUOIA：Developer Tools 2.0

Copilot has caught lightning in a bottle. GitHub’s AI-powered companion for developers, suggesting the code you might want to write next, is a hit in more ways than one. It’s beloved by engineers, attracting more than a million within about a year and delivering a substantial boost in productivity. (Even if you’re a top engineer, it might write half your code or more.) It also stands to become a commercial blockbuster. Charging $10-19 per seat per month, Copilot could grow to generate $1 billion or more in annual revenue among GitHub’s 100 million users.

Copilot’s success has set off a gold rush. Founders have raced to bring the power of large language models to all kinds of other industries, building tools to help professionals write, code, design and create media. There’s “Copilot for lawyers,” “Copilot for doctors” and “Copilot for designers,” and many more “Copilot for X”’s.

These are all exciting directions. We think generative AI stands to transform one industry after another, making all kinds of professionals more effective at work and delighting waves of consumers.

But there’s a lot more to do for developers. Copilot, which leverages OpenAI’s model Codex, may be just the opening salvo in AI’s transformation of how software engineers work. Andrej Karpathy predicted in 2017 that neural networks would create a new generation of software, “Software 2.0,” and we may see the same reinvention of the tooling that helps people make software — a “Developer Tools 2.0.”

There are many opportunities here. Some founders are iterating on the in-editor, get-help-while-you’re-coding experience Copilot popularized, trying different interaction patterns or different models. Think of Replit’s Ghostwriter, Soucegraph’s Cody, TabNine and others.

Many other opportunities lie further afield. You might target work engineers do beyond writing code, like debugging and documentation—or other work engineering orgs do, like incident response. You might think of value props other than “write code faster,” like “write code that’s more performant or more secure.” You might throw out the plugin form factor and rebuild an entire application. You might focus on personas other than the software engineer, like the data scientist who needs a boost writing in notebooks (see: Hex), or the data analyst toiling away writing SQL queries. There’s a wide waterfront of opportunities to explore and many places to hook into developers’ workflows.

Building legendary companies to enable this won’t be easy. Copilot faces legal scrutiny over concerns related to software piracy. Microsoft, the incumbent with control over both GitHub and VS Code, enjoys significant distribution advantages. Many founders have started building with LLMs, making many opportunities competitive. GitHub itself just announced plans to offer broader AI functionality through a brand-new version of Copilot, powered by GPT-4.

But in our view, there’s a massive opportunity for AI to transform software engineering, and it’s a matter of who, not if. We think important ingredients in who prevails will be focusing on developer experience, providing net-new capabilities, and making strategic choices around how to land and expand in developers’ workflows. Success here can mean rewriting how engineering happens—and having a chance at building a generational company.

机翻：
副驾驶在瓶子里发现了闪电。GitHub 的 AI 驱动的开发人员伴侣，可以建议您接下来可能要编写的代码，在很多方面都很受欢迎。它深受工程师的喜爱，在大约一年内吸引了超过一百万的人，并显着提高了生产力。（即使你是顶级工程师，它也可能会写出你一半或更多的代码。）它也有望成为商业大片。Copilot 每个席位每月收费 10-19 美元，可以在 GitHub 的1 亿用户中产生 10 亿美元或更多的年收入。

Copilot的成功掀起了淘金热。创始人竞相将大型语言模型的强大功能带到其他各种行业，构建工具来帮助专业人士编写、编码、设计和创建媒体。有“律师的 Copilot”、“医生的 Copilot”和“设计师的 Copilot”，还有更多“X 的 Copilot”。

这些都是令人兴奋的方向。我们认为生成式 AI将改变一个又一个行业，使各类专业人士的工作效率更高，并取悦一波又一波的消费者。

但是对于开发人员来说还有很多事情要做。利用 OpenAI 模型 Codex 的 Copilot 可能只是人工智能对软件工程师工作方式转变的开端。Andrej Karpathy在 2017 年预测，神经网络将创造新一代软件，即“软件 2.0”，我们可能会看到帮助人们制作软件的工具的同样改造——“开发者工具 2.0”。

这里有很多机会。一些创始人正在迭代 Copilot 普及的编辑 器内编码体验，尝试不同的交互模式或不同的模型。想想 Replit 的Ghostwriter、Soucegraph 的Cody、TabNine等。

许多其他机会在更远的地方。你可能会针对工程师在编写代码之外所做的工作，如调试和文档——或者其他工程组织所做的工作，如事件响应。您可能会想到“编写代码更快”以外的价值支持，例如“编写性能更高或更安全的代码”。您可能会丢弃插件形式因素并重建整个应用程序。您可能会关注软件工程师以外的角色，例如需要提高笔记本写作能力的数据科学家（请参阅：Hex），或者辛苦编写 SQL 查询的数据分析师。有广阔的探索机会和许多地方可以进入开发人员的工作流程。

建立传奇公司来实现这一点并不容易。Copilot因软件盗版问题而面临法律审查。微软是 GitHub 和 VS Code 的控制者，享有显着的分销优势。许多创始人已经开始使用法学硕士进行建设，这使得许多机会具有竞争力。GitHub 本身刚刚宣布计划通过由 GPT-4 提供支持的全新版本的 Copilot 提供更广泛的 AI 功能。

但在我们看来，AI 有巨大的机会来改变软件工程，这只是谁的问题，而不是是否。我们认为决定谁占上风的重要因素将是关注开发人员体验、提供全新功能以及围绕如何在开发人员的工作流程中着陆和扩展做出战略选择。此处的成功可能意味着重写工程的发生方式——并有机会建立一家世代相传的公司。