We have been operating a site called AI Coding.Info since July 2025.
This is a site that observes usage trends related to AI Coding Agents such as Claude Code, Codex CLI, Github Copilot, and Gemini 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.
https://qiita.com/kotauchisunsun/items/7b32a5f969a6322d66b4
AI Coding Agent repository utilization rate at the end of December was 5.3%, an increase of 0.5 percentage points from 4.8% last time. The upward trend remains unchanged. That's about it. In last month's article, we estimated the overall growth rate to be roughly less than 0.73%. Therefore, an increase of 0.5% can be considered a normal growth.

https://ai-coding.info/ja/agents
The share by product is as follows.
| Ranking | Product name | Share rate |
|---|---|---|
| 1st place | Claude Code | 29.1 % |
| 2nd place | Codex CLI | 25.2 % |
| 3rd place | Copilot Agent | 25.1 % |
| 4th place | Cursor | 11.0 % |
| 5th place | Gemini CLI | 6.6 % |
**Codex CLI has exceeded the share of Github Copilot Agent. ** In last month's article,
ClaudeCode V.S. While 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.
As I wrote above, the market share of Codex CLI has steadily increased over Copilot Agent. ** However, it is a very close rivalry. That's what it feels like. Referring to the data from 12/31, the number of repositories used for Copilot Agent is 175 and Codex CLI is 176, so there is only one difference. Therefore, in the current situation, **Codex CLI is just barely better. ** level, but by January 2026, the difference may have become larger. The reason is that the number of repositories used by Codex CLI increases faster than Copilot Agent, and we believe that the difference will increase next month.

https://ai-coding.info/ja/agents
The programming language in which AI Coding Agent is used most is "TypeScript," the second is "Python," the third is "Rust," the fourth is "Go," and the fifth is "C#." Up until now, I had considered the TOP3 to be one group, but I feel that it is starting to stabilize to 1st to 5th place. My evaluation of Go has changed over the past year, and my [July opinion] (https://qiita.com/kotauchisunsun/items/092784402a36d5705853) was that ``AI Coding Agent has not been implemented very well in Go.'' However, from [around November] (https://qiita.com/kotauchisunsun/items/7b32a5f969a6322d66b4), it seems to be starting to get closer to Rust, and is being actively adopted in OSS as well. Originally, the number of repositories used was about the same size as C#, but it feels like it is one step ahead of C#.

https://ai-coding.info/ja/agents
Recently, I have made the following announcements.
In this slide, we made predictions using data from July 1st to December 17th, when the service started, so this time let's look at data from July 1st to December 31st. That's what it means. This time we will also try fitting with a logistic function. (The term t_0 has been increased compared to the previous one)
First, let's fit the trends in the usage rate of AI Coding Agent across all programming languages.

The result is as follows.
K = 7.7575e-02
A = 2.8346e+00
B = 9.6907e-03
t0 = -4.4439e+00
決定係数 (R^2) = 0.9852
σ: 0.0012
I think that the coefficient of determination is 0.9852, which is a reasonably reliable range. In the data, there is a wave in the elapsed time on the horizontal axis between 25 and 50 days. This is due to changes in the scope of crawling and an increase in the types of AI Coding Agents. From now on, we will draw a forecast line until the end of 2026.

Therefore, the AI Coding Agent usage rate for all programming languages at the end of 2026 can be predicted to be 7.29-8.01%**. The parameter K represents the upper limit, which is 7.7575e-02, so the usage rate of all programming languages will reach saturation at the end of 2026. ** This seems to be a correct understanding.
Next, we will do the same fitting for TypeScript.

The result is as follows.
K = 1.0000e+00
A = 4.2080e+00
B = 3.8718e-03
t0 = 1.4383e+01
決定係数 (R^2) = 0.9199
σ: 0.0111
The coefficient of determination is 0.9199, so you can trust this to some extent. From here, we predict the AI Coding Agent usage rate at the end of 2026.

At the end of December 2026, the AI Coding Agent usage rate for TypeScript projects will be 61.81-68.47%. In fact, those who do not use AI Coding Agent are in the minority**. If you really trust this regression formula, it is K=1.0000e+00. Considering this, In the future, the usage rate of AI Coding Agent will be 100% in TypeScript projects The future can also be seen from mathematical models.
Here, we will also take a look at the languages that rank second or lower in the usage rate of AI Coding Agent: Python,'' Rust,'' Go,'' and C#.'' This is simplified and only provides long-term predictions.




Below is a table including TypeScript parameters.
| Language | K | A | B | t_0 | R^2 | σ |
|---|---|---|---|---|---|---|
| TypeScript | 1.0000e+00 | 4.2080e+00 | 3.8718e-03 | 1.4383e+01 | 0.9199 | 0.0111 |
| Python | 1.0000e+00 | 6.9301e+00 | 3.3556e-03 | 1.2202e+02 | 0.6722 | 0.0124 |
| Rust | 1.3049e-01 | 1.0538e+00 | 2.3307e-02 | 8.6224e-01 | 0.9652 | 0.0037 |
| Go | 1.2430e-01 | 5.8671e+00 | 2.2687e-02 | 1.7636e+00 | 0.9796 | 0.0045 |
| C# | 1.2233e-01 | 1.2652e+00 | 1.7884e-02 | 1.9869e+01 | 0.9686 | 0.0039 |
**TypeScript and Python can see a future where the usage rate of AI Coding Agent is 100%. ** This can be seen from the value of K. Regarding Python, the coefficient of determination is a low 0.672, so this cannot be said to be certain, but at this point, it is possible to say that this is the prediction. On the other hand, it can be said that TypeScript is more likely to increase the usage rate of AI Coding Agent than Python.
On the other hand, the upper limit of the usage rate of AI Coding Agents for Rust, Go, and C# is around 13%, which is currently reaching saturation. **. TypeScript and Python still had graphs that seemed to grow exponentially, but it is predicted that the usage rates of these languages are already reaching saturation. Rust is a language with a high rate of AI Coding Agent usage compared to the number of users. I had previously expressed the opinion. That's true, but among Rust language users, the number of people who use AI Coding Agent is actually a certain number, and that number is reaching saturation. You may be able to understand that.
While there is a view that TypeScript and Python's usage rates are skyrocketing, and the usage rate of AI Coding Agent will reach 100%!!'', the other opinion, As for Rust, Go, and C#, the usage rate of AI Coding Agents is reaching saturation'' is a consistent answer overall. First, the penetration rate of AI Coding Agents across all programming languages was 7.29% to 8.01% by the end of 2026, and the shape of the graph also showed a tendency for saturation. If the prediction results were such that the usage rate would be 100% for all programming languages, it would be logically inconsistent for the usage rate of all programming languages to fall between 7.29 and 8.01%. In that sense, even in Rust, Go, and C#, which currently have high penetration rates, the penetration rate of AI Coding Agents is about 13%, which is reaching saturation. This is because the prevalence of AI Coding Agents in all programming languages will not increase. This could be one of the explanations.
I think this is another way of expressing the fact that, as I've been saying for a while, there is uneven usage of AI Coding Agents between programming languages. This may be the opposite opinion, and the usage rate of AI Coding Agent among TypeScript and Python users is too high. Usage is spreading too quickly. ** This may be the correct understanding at present.
Originally, AI Coding.info started crawling around the middle of June, and the official release was on July 1st. Around that time, I participated in an AI-driven development event and gave a presentation on another topic.
https://aid.connpass.com/event/360978/
I was uploading the information of the dead slide at that time to X.
https://x.com/kotauchisunsun/status/1938218602563510494


In the last slide, I talked about my subjective view of the history of AI Coding at the time, and looking at it, it seems quite different from the current situation. That's the impression I get. I've been using RooCode since January 2025, and it's still used a lot locally. Part of it uses Gemini CLI. I haven't used OpenHands recently. As such, there was a lot of switching between tools, and it was extremely difficult to know which AI Coding Agent was best. That was the beginning of AI Coding.Info. Well, it wasn't that easy, and there were times when I saw hell (https://qiita.com/kotauchisunsun/items/a6ad9e7c3f6a7fc3ba1e), but in a sense, I was able to look at various AI Coding Agents from a calm perspective, and I didn't panic so much anymore. That's my impression. In that sense, I feel like I'm no longer swayed by AI-promoting information on SNS. And it feels much better than the shaky feeling at the beginning of 2025. Still, Gemini and Codex have come and gone, but in that sense, there was a lot of excitement around AI Coding as well. That's what it felt like. So, I would like to conclude with the trends in AI Coding in 2025. Good luck in 2025. What will happen in 2026? ? AI coding! ?
https://qiita.com/kotauchisunsun/items/7b32a5f969a6322d66b4
https://qiita.com/kotauchisunsun/items/e294d2cf419c46c02046
https://qiita.com/kotauchisunsun/items/5e5244fdabf577d7c879