Is Codex CLI making progress? ~ Summary of AI Coding trends in August 2025 as seen through data ~
AI Coding Agent trends in August
I have been running a site called AI Coding.Info since July.
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.
Change in survey method
We published an investigative article last month as well.
https://qiita.com/kotauchisunsun/items/092784402a36d5705853
This past month's update resulted in changes to the survey method.
- August 9, 2025 Increased crawl targets from TOP100 to TOP150
- August 10, 2025 Increased crawl targets from TOP150 to TOP200
- August 14, 2025 Increased crawl targets from TOP200 to TOP300
Previously, we were investigating 3,000 repositories, but this month that number has increased to 9,000. Due to some crawl errors, the numbers may differ from the actual numbers on the graph.
AI Coding Agent usage rate is 3.1%
AI Coding Agent repository usage rate at the end of August was 3.1%, an increase of 0.2 points from 2.9% last time. This is how many of the repositories we investigated showed evidence of the use of AI Coding Agent. This is the number. There were not many major movements during August. This time, the survey target has changed from OP 100 to TOP 300, and repositories with relatively few stars have also been included in the survey, but there is not much difference in actual usage. This is the result.
https://ai-coding.info/ja/agents
Share of AI Coding Agent by product
The share by product is as follows.
| Rank | Product name | Share rate |
|---|---|---|
| 1st place | Claude Code | 33.2% |
| 2nd place | Copilot Agent | 27.8% |
| 3rd place | Cursor | 16.6% |
| 4th place | Codex CLI | 11.5% |
| 5th place | Gemini CLI | 6.7% |
https://ai-coding.info/ja/agents
Looking at this month's data alone, in terms of usage share, Claude Code and Copilot Agent are in first place, followed by Cursor and Codex Agent, followed by Gemini CLI. That's my opinion. Compared to last month, Cursor's share has fallen significantly. However, this is largely due to the change in the survey method. That's my opinion. As mentioned earlier, Cursor's market share was certainly high when the survey was conducted on the TOP 100, but since the change to the TOP 300, the market share has been similar to the current one. As a result, repositories with higher star counts had more cursors, while repositories with lower star counts had fewer cursors. However, by expanding the scope of crawling, Claude Code and Copilot Agent, which were frequently used in lower-level repositories, expanded as a percentage of the total. That's my opinion. However, the growth of other AI Coding Agents is remarkable. That's a fact. For example, the number of downloads of Codex CLI has increased significantly, as shown in the tweet below. There is also data.
https://x.com/oikon48/status/1961996694146752770
Here, we will define and confirm the growth rate. This is calculated by dividing the number of repositories used on 9/1 by the number of repositories used on 8/18.
| Product name | Number of repositories used on 8/18 | Number of repositories used on 9/1 | Growth rate |
|---|---|---|---|
| Claude Code | 104 | 124 | 1.19 |
| Copilot Agent | 78 | 104 | 1.33 |
| Cursor | 59 | 62 | 1.05 |
| Codex CLI | 33 | 43 | 1.30 |
| Gemini CLI | 23 | 25 | 1.13 |
https://ai-coding.info/ja/agents
As you can see from this value, although **Claude Code and Copilot Agent have a larger number of repositories in use than Cursor as of August 18th, their growth rate is also higher. ** Therefore, under the current situation, it is difficult for Cursor to become the leading market share. Additionally, although Cursor has a higher number of repositories than Codex CLI and Gemini CLI, its growth rate is low, so it is possible that Cursor may drop in the rankings. **If we assume a fixed growth rate, Codex CLI may surpass Cursor in the number of repositories one month later in September. ** In the earlier tweet, we were talking based on the "number of npm downloads", but AI Coding.info is discussing based on "the number of repositories that contain the AGENTS.md file". Recently, there has been news that AGENTS.md is supported by various AI Coding Agents side by side. (As a result, the number of Codex CLI repositories included in AI Coding.info contains a slight error.)
https://gihyo.jp/article/2025/08/agents-md-site
**According to a survey on AI Coding.info, Codex CLI is already more used than Codex CLI and Gemini. ** However, is it really completely superior to Gemini? It's a very small difference, and I don't really feel like there's that much of a difference. This is related to the research method mentioned earlier, and since there are currently a large number of repositories in use for Copilot Agent and Claude Code, is Codex CLI growing at a rate that is faster than them? That being said, Claude Code and Copilot Agent are also growing fairly quickly, so the gap has not narrowed much. That's the impression I get.
https://ai-coding.info/ja/agents/codexcli
https://ai-coding.info/ja/agents/geminicli
AI Coding Agent usage status by programming language
**The programming language in which AI Coding Agent is used most is "Typescript," followed by "Rust" and "Python." ** The situation hasn't changed much between last month and this month, with Rust and Python swapping places from last month. 4th to 6th place hasn't changed much either. The rankings are C#, Go, and Ruby, but each has changed slightly. In that sense, there wasn't that much movement.
I think the crawl range has an impact here as well. Last month, there were 18 programming languages with at least one repository that was confirmed to be using the AI Coding Agent, but this month that number has increased to 24. In addition, there is a trend towards increasing use of AI Coding Agents in programming languages that originate from enterprises. For example, Kotlin, Dart, Swift. I feel that this language is difficult, and there is not much data that can be learned using OSS, so I think there is not enough data for AI to learn. That's what I thought. However, the use of AI seems to be progressing even in these fields. Personally, I find that PHP is also being used more and more in surprising places. Recently, I was talking with someone in the PHP world, and I heard that there was relatively little talk about AI at PHP conferences. I also write PHP, and I had a preconception that it would be relatively difficult for AI to output code if it was PHP like classic MPA (code that mixes html and <?php), but it seems that it is being used in some cases. (I don't think there is any code that mixes html and <?php like that these days.)
https://ai-coding.info/ja/languages
Monthly trends in the number of repositories used by AI Coding Agent
Let's take a look at the trends in August. Between August 9th and August 13th, the numbers changed significantly due to corrections to the crawl range. As we discussed in the previous chapter, the usage rate of Cursor was high before August 9th, but after the crawl range expanded after August 13th, we can see that the usage rate of Claude Code and Copilot Agent has increased. Additionally, the impact of the increase in crawl range is being seen here as well, and the number of repositories using Junie, which had previously been observed in small numbers, is also increasing. Recently, we have also been able to confirm repositories such as Trae IDE and Kiro, which are relatively popular.
https://ai-coding.info/ja/agents/junie
https://ai-coding.info/ja/agents/traeide
https://ai-coding.info/ja/agents/kiro
thoughts
Continuing from last monthThis is the second article. My feeling is that since Kiro came out at the end of last month, the numbers around here will increase. I thought so, but it didn't really affect OSS. However, Kiro has been adopted by an unexpected product, and is adopted by storybook, a relatively major OSS.
https://github.com/storybookjs/storybook
On the other hand, the announcement of GPT-5 and gpt-oss may also have an impact. I thought.
https://openai.com/ja-JP/gpt-5/
https://openai.com/ja-JP/index/introducing-gpt-oss/
Will the industry change drastically when such movements occur? ? That's what I thought when I watched it, but I wonder if it's actually that fast. There is also an impression that. Looking at it this way, you can see that Claude Code and Github Copilot have a surprisingly strong base. Also, it certainly seems that Codex is gaining momentum in terms of download numbers and search volumes, but on the other hand, there is a certain amount of latency before it spreads to OSS and becomes popular. There is also a way to recognize this. Therefore, what assumptions are made when looking at data and findings? I once again felt that this is extremely important.