GitHub 首席执行官 Thomas Dohmke 在最近的播客中讨论了 AI 编码工具的未来，他认为关键在于 AI 生成的代码能够让开发者快速修改，并能根据自身技能进行调整。Dohmke 强调，开发者应该能够根据自身需求，在 AI 生成的代码基础上进行快速修改，而不是花费时间引导 AI 进行更改。他认为，让开发者能够在不同工具之间切换，并选择最具 ROI 的工具，或者直接自己编写代码，是未来几年取得成功的关键。此外，Dohmke 还谈到了“氛围编码”现象，他认为初创公司不能仅仅依赖这种方式，非技术创始人如果没有开发者，将很难大规模构建初创企业。

💡 GitHub CEO 认为，AI 编码工具成功的关键在于，开发者能够快速修改 AI 生成的代码。这意味着开发者需要能够轻松地在本地机器上对 AI 生成的代码进行调整，而不是花费时间用自然语言描述他们已经知道如何用编程语言实现的功能。

🔑 Dohmke 强调，开发者应该能够自由地在 AI 生成的代码基础上进行修改，而不是浪费时间引导 AI 进行更改。他认为，这种方式可以提高生产力，而不是降低生产力。开发者应该能够选择最具 ROI 的工具，或者直接自己编写代码。

⚠️ Dohmke 认为，初创公司不能仅仅依赖“氛围编码”。他指出，非技术创始人如果没有开发者，将很难大规模构建初创企业。他认为，初创企业的价值不应该仅仅取决于使用廉价方式开发的东西。

🗣️ Dohmke 提到，开发者应该能够在 AI 辅助编写代码和自己编写代码之间自由切换。他认为，这种灵活性是未来几年 AI 编码工具成功的关键。

The CEO of GitHub described the "key for winning" when it comes to AI coding tools for software engineers.Thomas Dohmke said it's key for an AI agent to write code that the developer can quickly change using their own skills."Being able to pick the agent that provides the best ROI or do it themselves" is vital, he added.

AI coding tools are everywhere. So what's the approach that software engineers should be taking to get the most use out of them?

GitHub CEO Thomas Dohmke explained what he believes is the "key for winning" in this arena during an episode of "The MAD Podcast with Matt Turck" released Thursday.

"I think the key for winning is to have a continuum where you can have an agent write code for you and submit a pull request, but then if you as a developer see that code and you want to make three quick changes, you've got to be able to take that onto a local machine and just make those changes," Dohmke said.

The worst alternative, he said, is where a developer would be "trying to figure out how do I provide now feedback or prompt to describe in natural language where I already know how to do it in programming language."

Such a situation would be "basically replacing something that I can do in 3 seconds with something that might potentially take 3 minutes or even longer and that's obviously not more productivity, that's less productivity," Dohmke said.

In other words, software developers should be able to augment code generated by AI tools on the fly instead of wasting valuable time directing the AI to make such a change.

"Enabling developers to move between those categories and being able to pick the agent that provides the best ROI or do it themselves, I think that's the key for winning in the next few years," he concluded.

During a recent Q&A session at startup campus Station F at VivaTech in Paris, Dohmke also talked about the buzzy "vibe coding" phenomenon.

OpenAI cofounder Andrej Karpathy coined the term earlier this year to describe heavily leaning on AI tools to write code for you, allowing you to "fully give in to the vibes" and "forget the code even exists," in his words.

Dohmke said in the Q&A that he doesn't think startups can run on vibe coding alone.

A "non-technical founder will find it difficult to build a startup at scale without developers," he said, noting they "can't build a complex system to justify the next round."

"The value of your startup isn't defined by what you can develop using cheap measures," Dohmke said.