AI Is Starting to Help Build the AI That Comes Next

Artificial intelligence has spent the past few years transforming the way people write, code, search for information and analyze data. But a new development could be even more significant: AI systems are increasingly being used to help researchers build the next generation of AI systems.

Anthropic, the company behind the Claude family of AI models, recently revealed that Claude is now playing a substantial role in the company’s own artificial intelligence research and development. According to figures cited by Anthropic and analyzed by the nonprofit Epoch AI, Claude was involved in 26% of the company’s AI R&D work in August, compared with just 1% in March.

The development offers a glimpse into a future in which AI is not simply a tool used by software engineers, but an active collaborator in creating more capable AI.

From AI Assistant to AI Research Partner

Today’s AI coding assistants can already generate software, identify bugs, write tests and explain complicated code. However, helping to develop an AI model is considerably more complicated.

AI researchers work on tasks such as designing experiments, analyzing model behavior, improving training techniques, writing code and evaluating new systems. These activities require substantial reasoning and experimentation.

Anthropic says that more than 90% of its research involved some form of human-AI collaboration in August. The company also reported having around 30,000 AI agents operating on its internal platform, with human oversight built into the process.

This doesn’t mean Claude is independently designing and launching a new AI model from beginning to end. Human researchers remain responsible for directing the work and reviewing results.

Nevertheless, the increasing contribution of AI to AI development represents an important shift.

Why This Could Matter

One reason the development is attracting attention is the possibility of creating a feedback loop.

Researchers build an AI system. That system helps researchers discover better algorithms, write more efficient code or design improved experiments. Those improvements can then contribute to the development of a more capable AI system.

The new system could potentially become even more useful for research, creating another cycle of improvement.

This concept is sometimes described as recursive AI development or AI-assisted AI research. It does not necessarily mean that AI will suddenly become completely autonomous. Instead, it suggests that the amount of human work required for each generation of AI could gradually change.

The economic implications could also be significant. If AI researchers can delegate more routine experimental and engineering tasks to AI agents, small teams could potentially accomplish work that previously required much larger teams.

The Human Role Is Still Critical

Despite the headlines surrounding AI systems helping create their successors, there is an important distinction between AI-assisted development and autonomous self-improvement.

Anthropic has emphasized that Claude’s work remains under human supervision. The company’s agents are subject to controls, and actions are screened before they are carried out.

That distinction matters because AI models can still make incorrect assumptions, produce flawed code or generate convincing but inaccurate explanations.

In scientific research, an apparently successful experiment is not necessarily a meaningful discovery. Researchers need to determine whether the result can be reproduced, whether the methodology is sound and whether the conclusions actually follow from the evidence.

Human researchers therefore remain essential for setting goals, judging results and deciding which discoveries are worth pursuing.

A Growing Industry-Wide Trend

Anthropic’s announcement is part of a broader movement toward AI agents that can perform longer and more complicated sequences of tasks.

Technology companies are increasingly developing systems that can use software tools, interact with codebases, conduct research and coordinate multiple tasks rather than simply respond to individual questions.

Recent industry developments also show increasing interest in combining AI with robotics, advanced computing infrastructure and scientific research. McKinsey’s 2026 technology outlook, for example, highlights AI and other frontier technologies as major areas of technological development.

The direction is becoming clear: AI is moving from a system that primarily answers questions toward one that can increasingly perform work.

What Happens Next?

The most interesting question isn’t whether AI will completely replace human AI researchers. It is how the relationship between humans and AI researchers will evolve.

If AI systems become better at programming, experimentation and technical reasoning, researchers may spend less time performing individual tasks and more time deciding which problems deserve attention.

That could accelerate progress in fields far beyond artificial intelligence, including medicine, materials science, mathematics and engineering.

At the same time, greater AI involvement in AI development makes transparency and safety increasingly important. The more responsibility AI systems receive, the more important it becomes to understand what they are doing, verify their results and establish clear limits on their actions.

For now, humans are still firmly in the loop. But the direction of travel is notable.

AI is no longer just being built for people to use. It is increasingly being used to help build the technology itself.

And if that trend continues, the next major leap in artificial intelligence may not come solely from humans developing better AI. It could come from humans and AI working together to discover what comes next.