AI Coding Agent November trends

We have been operating a site called AI Coding.Info since July 2025.

https://ai-coding.info/ja

This is a site that observes usage trends related to AI Coding Agents such as Claude Code, Gemini, or Codex from a fixed point from information in Github repositories. To determine the use of AI Coding Agent, we conduct daily surveys under the following conditions.

• Daily inspection of 9,000 public repositories on Github
  • TOP 300 Github stars by programming language
  • Researched from 30 types of programming languages
• Targets 16 types of AI Coding Agents
• Only when the rule files used by various AI Coding Agents are in the Github repository, it is determined that AI Coding Agent is being used.

Past trends

https://qiita.com/kotauchisunsun/items/e294d2cf419c46c02046

https://qiita.com/kotauchisunsun/items/5e5244fdabf577d7c879

https://qiita.com/kotauchisunsun/items/a1e06dd590f945ae09ef

https://qiita.com/kotauchisunsun/items/092784402a36d5705853

AI Coding Agent usage rate is 4.8%

The AI ​​Coding Agent repository utilization rate at the end of November was 4.8%, an increase of 0.6 percentage points from 4.3% last time. In last month's report, we explained the trends in this usage rate.

In the previous article, we predicted that for all programming languages, AI Coding Agent usage would plateau at around 13% I predicted. Among them,

$$y(t) = \frac{K}{1 + Ae^{-Bt}}$$

When fitting the utilization rate with the model formula,

K = 1.2862e-01
A = 5.0299e+00
B = 7.5901e-03
決定係数 (R^2) = 0.9534
平均二乗誤差 (MSE) = 0.0000
二乗平均平方根誤差 (RMSE) = 0.0014

The value is now . Differentiating the original equation with respect to t, we get

$$\frac{dy}{dt} = \frac{A B K\, e^{-B t}} {\left( A e^{-B t} + 1 \right)^{2}}$$

Differentiating with respect to t again, we get

$$\frac{d^{2} y}{dt^{2}} = \frac{A B^{2} K (A - e^{B t}) e^{B t}} {A^{3} + 3 A^{2} e^{B t} + 3 A e^{2 B t} + e^{3 B t}}$$

It becomes. In order to find the inflection point of this, if we solve for t with the left side = 0, we get

$$t_0 = \frac{log(A)}{B}$$

It becomes. Finding the slope of y when $t_0$ is,

$$\left. \frac{dy}{dt} \right|_{t = t_0} = \frac{BK}{4} \approx 2.441 \times 10^{-4}$$

The previous formula calculated the amount per day, so to convert it to a monthly amount, multiply by 30. Then

$$2.441 \times 10^{-4} \times 30 \approx 0.007321$$

It becomes. Considering this, we can see that the growth rate of AI Coding Agent usage across programming languages ​​is less than approximately 0.73% per month. ** Considering this, the usage rate has increased by 0.6% this month. This is relatively within the expected range. ** That means.

https://ai-coding.info/ja/agents

Share of AI Coding Agent by product

The share by product is as follows.

Ranking	Product name	Share rate
1st place	Claude Code	28.6 %
2nd place	Copilot Agent	26.8 %
3rd place	Codex CLI	22.5 %
4th place	Cursor	11.6 %
5th place	Gemini CLI	6.5 %

As was the case last month, I feel that the share of Codex CLI is gradually increasing. Actually, there hasn't been much change in terms of Gemini CLI. I have the impression that the market share of Cursor and ClaudeCode has dropped significantly. (Cursor actually supports AGNETS.md, so it is difficult to say whether the market share is really declining.) How do you view this composition? However, ClaudeCode does not support AGENTS.md. I think this serves as a bulwark to prevent tool changes (https://qiita.com/kotauchisunsun/items/5e5244fdabf577d7c879). However, it seems that the outflow is still progressing. That's what I think. I have the impression that AGENTS.md, such as Codex CLI, is eating up ClaudeCode's market share, while Github Copilot's market share is unlikely to decline due to the strength of the VSCode standard. Perhaps, while ClaudeCode V.S. AGENTS.md (Codex CLI) is competing for market share, I can see a future in which Github Copilot, which has little leakage, will take first place in market share. That's what I feel. **

https://ai-coding.info/ja/agents

AI Coding Agent usage status by programming language

The programming language in which AI Coding Agent is used most is "TypeScript," the second is "Python," and the third is "Rust." The trend here remains almost unchanged. The second group is Go, C#, Kotlin, and Ruby. In this area, the rankings have been changing for several months. However, this month, Go has surprisingly caught up with Rust. There is a feeling.

https://ai-coding.info/ja/agents

Up until now, I have written that AI Coding Agent may not be very useful in Go. This may be due to the lack of use of CNCF-related repositories. I wrote about this before, but now that I look at it, I think it's being used more and more in such infrastructure fields.

For example, in helm there is an adoption of AGENTS.md.

https://github.com/helm/helm

vittess has adopted CLAUDE.md and AGENTS.md.

https://github.com/vitessio/vitess

Recently, I often hear about TiDB in the context of large-scale MySQL, and I can see settings for AGENTS.md, CLAUDE.md, and GithubCopilot.

https://github.com/pingcap/tidb

https://ai-coding.info/ja/languages/go

Monthly trends in the number of repositories used by AI Coding Agent

As of December 1st, the total number of AI Coding Agent repositories used was 619. Last time there were 511 cases, so this is an increase of 108 cases. However, I think it's a little different. There is a recognition that

Actually, in the previous chapter, I wrote that the usage rate of AI Coding Agent was 4.8%. At this time, the number of repositories is also reported within the service, but the number of repositories calculated based on the usage rate of AI Coding Agent is 431. Actually, it is different from 619 above. ** This is a slightly confusing difference in aggregation rules. I think you can understand by giving an example.

Repository	Claude Code	Github Copilot	Codex CLI
A	×	×	×
B	〇	×	×
C	×	〇	×
D	×	〇	〇

As mentioned above, there are four repositories from A to D. **A does not use AI Coding Agent, so the usage rate is 3/4=75%. And the number of AI Coding Agents used based on usage rate is 3. However, all the stacked graphs in this section are counted separately. Therefore, the number of repositories used is 4 (D is two for GithubCopilot and CodexCLI). ** When aggregating based on usage rate, the count will only increase by 1 even if multiple AI Coding Agents are used in one repository, but with the aggregation method in this section, if multiple AI Coding Agents are used, the count will increase by the number of AI Coding Agents. What does this mean? So, if you divide the 619 counted in this section by the 431 counted based on the usage rate, you will find that the number of AI Coding Agents used per repository is 1.50. ** In fact, repositories that use **AI Coding Agent tend to already use multiple AI Coding Agents. **

thoughts

This time, hybrid use of AI Coding Agent. I wrote about trends regarding repositories that use multiple AI Coding Agents. I review the data in this way once a month, and this time I find the movement related to Go disgusting. That was the beginning. As I wrote in the article, Go has surprisingly few AI Coding Agent users. That's what I was aware of, but this time, I had the impression that it was rapidly growing and catching up with Rust. So, let's dig a little deeper. I thought. After looking into several famous repositories, I found settings for multiple AI Coding Agents, so I wonder if there is a trend towards using multiple AI Coding Agents in earnest. This feeling has become stronger. I have explained AGENTS.md in the past, but as the trend towards common rule files emerges, there will be many cases where multiple rule files coexist, such as CLAUDECODE.md and AGENTS.md. There was a prediction. However, I was aware that this was a problem when AGENTS.md's share increased, but since around October, it has been listed in the top 3, so I feel that it has become a presence that cannot be ignored. And when I actually calculated it, it turned out to be more than I expected. That was my impression. In this article, to put it simply, if you take out two repositories that use AI Coding Agent, there will be three rule files. In this situation, there may be quite a few. It felt like that. When you think about it, everyone places a lot of importance on the rule file and wishes to switch over and standardize it. That's strong. I felt that again.