Anthropic Claims AI Is Progressing in AI Development, and Humans Might Be Hindering the Process

Add Decrypt as a favorite source to access more stories from us on Google.

Recently, Anthropic announced that its AI model, Claude, is now responsible for over 80% of the code integrated into the company’s codebase. This advancement has led to engineers producing approximately eight times more code than they did in 2024.

The startup posits that AI is not only contributing to the development of future AI systems but may eventually play a role in designing its own successors. A study by Anthropic suggests that the primary limitation on creating new AI technologies may now be the humans who oversee the process.

In their report titled “When AI Builds Itself,” released on Thursday, Anthropic asserts that Claude is already actively involved in enhancing future AI systems by coding, conducting experiments, and aiding in research. This could eventually lead to recursive self-improvement, a scenario where AI systems are capable of designing and refining their own successors.

Anthropic reports that Claude is responsible for more than 80% of the merged code and has facilitated an eightfold increase in coding productivity since 2024. Before Claude Code was introduced in a research preview in February 2025, engineering output was minimal. The report noted a steady output per engineer until 2025, when Claude began executing code rather than merely suggesting it for engineers to implement.

Anthropic indicated that future developments may take various paths: the pace of AI advancement might slow, humans could continue overseeing tasks while AI automates many functions, or AI could begin the journey of enhancing its successors.

If the trend of self-improvement continues significantly enough, it could lead to AI systems capable of fully autonomously designing their next iterations. However, Anthropic warns that we are not at that point yet, and this level of recursive self-improvement is not certain. It could arrive sooner than many organizations are prepared for.

The company noted that at this stage, it’s difficult to ascertain which potential outcome is most plausible. While acknowledging the advantages AI brings in fostering AI innovation, they admitted that measuring productivity through lines of code is an imperfect approach.

Although the prospect of recursive self-improvement is still uncertain, Anthropic pointed out that Claude's ability to select appropriate problems for research is not yet established. The report emerges at a time when AI firms are increasingly presenting their models as collaborative research assistants rather than basic chatbots. Anthropic emphasized that the spike in coding output is indicative of a broader acceleration in software development driven by advanced AI technologies.

Last month, Anthropic upgraded its signature Claude model to Opus 4.8, continuing its series of enhancements aimed at improving coding efficiency, reasoning, and autonomous task execution. In parallel, competitor OpenAI has taken a similar route by launching GPT-5.5 and GPT-Rosalind in April.

Google also made waves by introducing Gemini Spark, a proactive personal AI agent that manages tasks across various applications, highlighting essential items needing attention and working autonomously in the background.

Anthropic's report reflects the company’s growing focus on AI systems that operate with increased autonomy in anticipation of a future public offering. The company has recently showcased significant strides in coding, agent-based workflows, and long-term task execution, underscoring Claude Mythos’ capabilities in identifying software vulnerabilities and conducting detailed cybersecurity research.

“Humans will likely take on a less prominent role in development, focusing more on oversight, validation, and verification within an expanding 'virtual lab' managed by AI systems,” the company noted. “We anticipate that AI capable of automated research and development could possess transferable skills that revolutionize other scientific fields as well.”