On GitHub, humans can no longer compete with AI.

If someone had told a programmer a few years ago, "You might have to compete with AI for GitHub commit records in the future," he would probably have laughed out loud.

But now, he might not be able to laugh at all.

According to the latest analysis report released by SemiAnalysis, Anthropic's Claude Code has currently contributed 4% of the public commits on GitHub, and is expected to:

Reach 20% of the daily commit volume by the end of 2026.

This is not just a simple numbers game.

When an AI tool starts to "make its presence felt" on the world's largest code hosting platform, it is actually redefining what it means to "write code".

01

AI "Dominates" GitHub

4% may not seem like a large number, but what's scary is the meaning behind this number.

The daily commit volume on GitHub is an astronomical figure. Tens of millions of programmers around the world push code, fix bugs, and release new features on this platform. And now, 1 out of every 25 commits comes from AI.

Boris Cherny, the person in charge of Anthropic Claude Code, openly "showed off" on X: His team now uses Claude Code 100% to write code, and no longer manually performs even small edits.

Even more incredibly, they built the Cowork application using Claude Code in just a week and a half.

This improvement in efficiency is not linear, but exponential.

But what really shocks people is not the speed, but the quality. A corporate user revealed that he spends 80% of his time using Claude Code and the remaining 20% using other tools.

"My company pays for Claude Code, and I don't even look at the cost."

This statement is quite interesting - when a tool is so good that people "don't look at the cost," it means that the value it creates far exceeds the price.

An industry insider once evaluated the advantages of AI Coding like this: "AI can bypass bureaucracy. If indecision paralyzes large organizations, AI doesn't care. It will happily generate a Version 1."

This statement points out the core advantage of AI programming - no baggage, no hesitation, and no "perfectionism anxiety."

02

The "Existential Crisis" of Programmers

But every coin has two sides.

On Hacker News, a user shared his frustration: "Many times I wanted the code to look a certain way, but it kept reverting to the way it wanted to do things... Eventually, I found it easier not to fight with it and let it do things its way."

This passage reveals a subtle power shift:

From "humans guiding AI" to "humans adapting to AI".

Dan Shipper, the CEO of Every Company, wrote in his blog: "We are in a new era of autonomous programming. You can build amazing and complex applications without looking at a single line of code."

It sounds great, but it also means that the traditional "programmer" is disappearing.

If you can build an application without looking at the code, is "being able to write code" still a core skill?

Analysts at SemiAnalysis predict that this trend will drive Anthropic to achieve explosive growth in 2026, even surpassing OpenAI. In contrast, although GitHub Copilot and Office Copilot had a one - year head start, "they have made almost no progress as products."

This comparison is cruel but also very telling: in the AI era, the first - mover advantage may not be as important as the product experience.

03

Redefining "Programmers"

But programmers probably don't need to be overly anxious. The programmer position will not disappear; it's just that the definition of this profession is changing.

As Dan Shipper said, even in 2025, "you still need to truly understand the underlying architecture, and maybe you still need to look at the code."

But the meaning of this "need" has changed.

Programmers are changing from "code writers" to "AI coordinators".

You need to know how to communicate with AI, how to review its output, and how to correct it when it makes mistakes. You need to understand the system architecture, but you don't necessarily need to implement every line of code yourself.

The reflection of a Google engineer is quite representative: The community's discussion about AI programming ability is "tense." On the one hand, people are amazed at the improvement in ability, and on the other hand, they are worried about being replaced. But he emphasized that domain expertise is still important, and the gap between prototypes and production environments still exists.

When AI programming is cheap enough and good enough, the economics of the entire software development will change.

Maybe 20% of the GitHub commit volume is just the beginning. Maybe in a few years, we will see 50%, 80%, or even a higher proportion coming from AI.

This doesn't necessarily mean the end, but a new beginning. Real programmers will not be replaced by AI, but will learn how to make AI their most powerful tool.

Just as calculators didn't make mathematicians unemployed, AI won't make programmers unemployed - it will only make those who refuse to evolve unemployed.

Code is the machine language and the native language of AI. Returning the machine language to the machine itself and having humans describe ideas in natural language seems to be a more natural outcome.

This article is from the WeChat official account "GeekPark" (ID: geekpark). Author: Hua Lin Wuwang, Editor: Jing Yu. Republished by 36Kr with permission.