Jensen Huang has been saying repeatedly over the past two years that "the age of agentic AI is here" and that it's a "trillion-dollar opportunity." Last October he said every Nvidia engineer uses Cursor, and at this year's GTC he painted a picture of 75,000 people working alongside 7.5 million agents.
It sounds like coding agents are already everywhere. I checked the actual numbers.
The Numbers Are Surprising
Claude.ai has roughly 10–20 million monthly active users. Third-party estimates put Claude Code at around 1.6 million weekly active users. Cursor has over 2 million users, with 1 million paying. On the open-source side, OpenCode has 140,000 stars on GitHub, and Cline has been installed more than 5 million times in VS Code.
Those numbers aren't small. But GitHub Copilot alone has nearly 20 million users, and a JetBrains survey early this year found that 74% of developers already use some kind of AI coding tool.
Copilot is mostly autocomplete; it doesn't really count as an agent. True agentic tools—the ones where you give a task and they read files, write code, and run tests on their own—have somewhere between a few million and a little over ten million users combined.
Something called "world-changing" by the highest-valued company on Earth has a user base at that level. I would have guessed at least tens of millions.
Domestically (in China), things are more lively. The Kimi platform has over 30 million MAU, and ByteDance's Trae has more than 6 million registrations. But those figures include plenty of non-programming use cases; the number of people actually using coding agents isn't that high.
A lot of people are discussing how to configure Skills or connect MCPs. But maybe we should take a step back: why can't so many people even take the first step?
It's Not Just About Writing Code
The most common misconception is treating a coding agent as a "tool for writing code." Non-coders think it's irrelevant; coders think it's just an upgraded Copilot.
In reality, it does far more than code. Splicing videos, batch processing files, scraping data from websites, debugging unfamiliar software—it can handle all of that. Simply put, it helps you control your computer to get tasks done; code is just its operating language. I wrote before that this is a meta-ability—letting ordinary people truly control their computers with an extremely low barrier to entry.
Phones are the exception. Phones are inherently GUI-centric, so coding agents don't work well on them. Those projects using models to control phones were trendy for a while, but the approach is completely different and not quite there yet.
The NPC Woke Up
The deeper problem is mindset.
When it comes to using software, people are used to preset interactions. Click here for this, drag there for that—everything has been designed in advance. NPCs in games work the same way: they give you three options to choose from, and it doesn't really matter whether you read the dialogue.
Now the NPC can suddenly think. It waits for you to speak, then does what you ask.
This feeling is the same as when ChatGPT first came out. I made a lot of tutorials teaching people how to use it, and later found that most people were stuck on expression. They felt they had to become "prompt engineers"—speak precisely, make the AI obey, and at an advanced level do some tricks.
It's not that complicated. Just treat the coding agent like a colleague. It understands what you say and can read through the files in your project. Explain clearly what you want to do, and you're mostly done.
Don't Do It Yourself
There's another pitfall related to habit.
The more capable you are, the easier it is to fall into it. When facing a problem, the first instinct is to do it yourself. It's like being a manager—someone else can clearly do the work, but you always feel faster doing it yourself.
But a coding agent might be ten times faster than you. The quality may be temporarily lower, but iteration speed makes up for it. The problem is, once you start doing it yourself, you slide back into the old path—ask DeepSeek or Doubao for help, make the changes yourself, use AI as an advisor. The work is still yours.
I now delegate 95% of my computer work to coding agents. Research, writing, programming, emailing, operating web pages. Once you cross over, you don't need anyone to teach you, because you can have it help you figure out how to use it better. That's the meta-ability. That's how I built this website from scratch—no front-end knowledge, no DNS knowledge, no SEO knowledge; I let Claude Code handle it all.
Human in the Loop
But don't be too optimistic either.
Concepts like agent managers and multi-agent orchestration sound beautiful—let AI manage AI, fully automated. The direction is right, but we're not there yet.
In practice, having a knowledgeable human in the loop makes a huge difference in efficiency. If you completely let agents orchestrate themselves, things fall apart once tasks get complex. I talked about the Worker and Manager relationship before—this is exactly that.
In the short term, someone still needs to be in the middle. But that person's role isn't to do the work with their own hands; it's to articulate what they want, glance at whether the result is right, and make the call at key moments.
Once you cross this step, many things naturally fall into place. If you can't cross it, you'll keep watching others use it from the outside.