Can AI Help Build the Next AI? Anthropic's New Claude Report Explained
What Is Anthropic's New Report About?
AI is usually known for answering questions, writing text, creating images, and helping people write code.
But AI is now taking on a larger role in AI research and development.
On September 17, 2026, Anthropic shared new information about how much Claude contributes to its own AI development work.
According to the company, Claude was leading about 26% of its measured AI R&D work in August 2026. Anthropic also reported that more than 90% of the measured work involved AI collaboration or a higher level of AI involvement.
So, is Claude now building the next AI by itself?
No.
These figures show that Claude is taking on a growing share of research and engineering tasks, while humans continue to provide direction and oversight.
What Did Anthropic Announce?
Anthropic introduced a new way to measure how much AI helps with AI research and development.
Simply saying that AI is helping researchers does not tell us how much work AI is actually doing.
For some tasks, AI may do most of the work while a human gives instructions and checks the results.
How Much Work Is Claude Doing?
According to Anthropic, Claude was leading about 26% of the measured AI R&D work in August 2026.
→ Source: Anthropic — Measurements for understanding the pace of AI development
The word "leading" is important.
It does not mean Claude is independently running Anthropic's AI research.
It also does not mean that Claude built 26% of a new Claude model by itself.
Instead, Anthropic uses "AI leads" for work where the AI can perform most of the task from a relatively simple, high-level instruction, while humans still provide oversight.
This means Claude is becoming capable of handling larger parts of some research and engineering tasks.
What Does "AI Leads" Mean?
Anthropic uses several levels to describe how much AI is involved in a task.
The levels range from little or no meaningful AI involvement to highly independent AI work.
One important level is AL3, or "AI collaborates."
At this level, AI works together with a human and can complete important parts of a task.
Another level is AL4, or "AI leads."
At this level, AI can complete most of the work after receiving a high-level instruction. Humans still supervise the work and remain responsible for the overall process.
Anthropic also defines AL5 as fully autonomous work.
The important point is that Anthropic's current report should not be read as saying that Claude has reached fully autonomous AI development.
What Kind of Work Can Claude Do?
 |
| Figure 1. AI-assisted research and engineering workflow |
Claude is being used for real research and engineering work.
This can include tasks such as:
- Writing and changing code
- Testing ideas
- Running experiments
- Looking at results
- Solving technical problems
- Improving technical systems
This is different from simply asking an AI chatbot a question.
In these situations, AI can become part of a longer work process.
For example, a researcher can give Claude a goal. Claude can then work through several steps, test changes, and produce results that a human can review.
This allows AI to take on a larger part of the research process.
A Real Example: AI Helping With Scientific Research
Anthropic also shared a specific example involving biomolecular modeling.
Researchers used Claude to help work on and improve more than 30 biomolecular modeling models.
According to Anthropic, the work took about four weeks, and the models were made about four times faster on average.
→ Source: Anthropic — How Claude is uplifting biomolecular modeling
This example is useful because it shows what AI involvement can look like in real research.
Claude was not simply answering questions about science.
It was used to help improve technical models and work through an engineering problem.
This does not mean that AI can replace scientists.
Instead, it shows how an AI model can become a useful part of a research team.
What Are the 30,000 AI Agents?
Anthropic also reported that about 30,000 AI agents have been used through one of its internal platforms for research and engineering work.
An AI agent can do more than provide a single answer.
It can perform a series of actions to work toward a goal.
However, the number 30,000 should not be understood as 30,000 AI employees.
These are AI agents operating through Anthropic's internal systems.
They are part of the company's research and engineering workflow.
How Are Humans Still Involved?
Even when AI performs a large part of a task, humans still have an important role.
Researchers can decide what problems to work on, give instructions, check results, and review the work done by AI.
Anthropic also described systems for monitoring AI agents.
These monitoring systems are designed to help detect potentially problematic actions while AI agents are working.
This becomes more important as AI systems are given more tools and are allowed to perform longer tasks.
The goal is not simply to give AI more freedom.
It is also important to understand what the AI is doing and when humans need to step in.
Does This Mean Claude Is Building Its Own Successor?
No.
This is one of the easiest parts of the news to misunderstand.
Claude is helping with some of the work involved in AI research and development.
But that does not mean Claude independently decided to create a new AI model and completed the entire process without humans.
Humans still provide goals, instructions, review, and oversight.
The 26% figure is also limited to the AI R&D work Anthropic measured.
It does not mean that 26% of everything Anthropic does is performed by Claude.
Why Is Anthropic Measuring This?
One important question is:
How much useful work can an AI model actually do?
Anthropic's new measurement system tries to answer this question for AI research and development.
Instead of only measuring what AI can do in individual tasks, the company is also looking at how much real work AI can perform as part of a research process.
This could become increasingly important as AI systems take on more complex tasks.
What Are the Limits of These Numbers?
The numbers in the report should be understood in context.
Anthropic created this measurement system to track its own AI R&D work.
There is currently no single worldwide standard for measuring how much AI contributes to AI research.
This means that the 26% figure should not automatically be compared with a percentage from another AI company unless the companies use the same measurement method.
The way a company defines and measures a task can also affect the final number.
Anthropic's report is therefore useful for showing how AI involvement is changing inside its own research environment, but it should not be treated as a perfect measurement of all AI development.
Why Does This Matter?
The interesting part of this news is not simply that Claude can write code.
AI has already been helping programmers and researchers for some time.
What is changing is the amount of work AI can handle inside a larger research process.
→ Source: Reuters — Anthropic says Claude now leads a quarter of work building its next AI models
AI can now help with coding, experiments, testing, analysis, and other technical tasks.
If this trend continues, AI could become an increasingly important tool for developing future AI systems.
At the same time, humans still need to set goals, check results, and decide how AI should be used.
So the bigger question is no longer just:
"Can AI help researchers?"
It is becoming:
"How much of the research work can AI do, and how should humans supervise it?"
What Should Beginners Know?
The main idea is simple.
AI is starting to take on more work in AI research and development.
Claude can handle parts of research and engineering tasks, but humans still set goals, review results, and supervise AI systems.
This does not mean that AI has reached fully autonomous AI development.
Key Takeaways
- Claude was leading about 26% of Anthropic's measured AI R&D work in August 2026.
- AI is taking on a growing role in AI research and engineering tasks.
- Anthropic reported using about 30,000 AI agents for research and engineering work.
- Humans are still involved in setting goals, reviewing results, and supervising AI systems.
FAQ
Q. Is Claude building the next AI by itself?
A. No. Claude is helping with AI research and development, but humans still provide direction, review, and oversight.
Q. What does the 26% figure mean?
A. It means Claude was leading about 26% of the AI R&D work measured by Anthropic in August 2026. It does not mean Claude built 26% of a new AI model by itself.
Q. What are the 30,000 AI agents?
A. Anthropic reported that about 30,000 AI agents have been used through an internal platform for research and engineering work. They are not 30,000 independent AI researchers.
Q. Can AI completely replace AI researchers?
A. Not based on this report. AI can handle a growing amount of research and engineering work, but humans still set goals, review results, and supervise AI systems.