Quick guide
Quick answer
Anthropic says Claude is writing much of the code used to build the next Claude, signaling faster AI development while human teams still set goals and review the work.
What you'll find here
- Start with the plain-English guide
Anthropic says Claude is writing much of the code used to build the next Claude, signaling faster AI development while human teams still set goals and review the work.
- Anthropic says more than 80% of merged code was authored by Claude as of May 2026; recursive self-improvement means AI helps develop its successor; humans still set goals and decide what matters; longer autonomous task performance is accelerating.
Here's a weird sentence that landed this week.
The company that builds Claude just said its own AI is writing most of the code that builds the next Claude.
Not a rumor. Anthropic published it on their site under the heading When AI builds itself. They call the end state of that trend recursive self-improvement. Fancy words for a simple idea:an AI that can design and develop its own successor with little or no human help.
They are clear about one thing. We are not there yet. Claude is not fully autonomous. Humans still set the goals and decide what matters. But the numbers they shared should make you sit up.
As of May 2026, Anthropic says more than 80% of the code they merge into their own codebase was authored by Claude. Before their coding agent launched in early 2025, that share was in the low single digits.
In the second quarter of 2026, the typical Anthropic engineer was merging about 8 times as much code per day as in 2024. Anthropic itself says lines of code is a messy measure. Quantity is not quality. Still, that is a lot of acceleration in one place.
They also point to a public trend:the length of tasks AI can finish reliably on its own has been roughly doubling every four months. A few years ago it was minutes. Then hours. Now they talk about multi-hour and even longer stretches of work.
Why should you care if you just use ChatGPT or Claude for emails, school help, or a messy inbox?
Because the everyday risk is not a movie villain. It is speed. When the people building these tools use the tools to build the next ones, the gap between “what I understand” and “what just shipped” gets wider. Your job, your kid's homework help, the apps you trust with your info. They all sit downstream of that speed.
Anthropic's own pitch is that the public, governments, and other labs need better visibility into how fast this is moving. They say recursive self-improvement could arrive sooner than most institutions are ready for. They also say a slowdown could be good if other frontier labs can be verified as slowing too. A pause by one company alone just hands the lead to someone else.
Look, I get it. This stuff can sound abstract. Translate it like this:the shop floor is starting to run itself, and the supervisors are still human. For now. That is useful to know before the next “everything is fine” press release.
You can read Anthropic's report here:When AI builds itself. Coverage from Euronews matches the same core story.
My ask if you use AI this week:treat confident answers like a draft, not a finished product. Ask where the number came from. Ask what it might have skipped. The people inside the labs are watching these systems closely for a reason. You can watch them a little closer too.
What part of this still feels fuzzy to you?
Joe Foley
Joe Foley is the creator of AI for Ordinary People. He helps beginners, parents, creators, and small business owners use AI in simple, practical ways. Joe has been podcasting since 2013 and creates plain-English guides, prompts, and workflows for people who want useful help without tech jargon. His goal is to make AI feel less confusing and more useful for real life, real work, and everyday decisions.