Tech • IA • Crypto
Anthropic just published "When AI Builds Itself": 80% of the next Claude’s code is generated by AI
9.3/10TL;DR
A new report from Anthropic indicates that over 80% of its latest AI system’s code is now generated by AI itself, highlighting a rapid shift toward self-improving systems.
KEY POINTS
AI now builds most of its own code
The latest generation of Claude is reportedly developed with more than 80% of its code written by AI systems, marking a major shift in software engineering. Just two years ago, AI contributions to its own development were minimal, underscoring the سرعت of this transition. The change accelerated significantly after the deployment of Claude Code in early 2025.
Explosive growth in autonomy
The system’s ability to complete tasks independently has expanded dramatically. In March 2024, it could operate autonomously for only 4 minutes per session. Within a year, this increased to 1.5 hours, and it now reaches up to 12 consecutive hours of uninterrupted work.
Doubling capability at a rapid pace
Operational autonomy is reportedly doubling at a consistent rate, suggesting exponential progress. If this trend continues, AI systems could soon handle workloads equivalent to a full human workweek without interruption as early as 2027.
Emergence of recursive self-improvement
The development process is entering a phase of recursive self-improvement, where each version of the AI contributes to building more advanced successors. This creates a feedback loop that accelerates innovation cycles beyond traditional human-led development timelines.
Limits remain in judgment and prioritization
Despite technical advances, current systems still lack human-like judgment, particularly in deciding what tasks matter most and when to stop. This limitation remains a critical barrier to fully autonomous decision-making.
Tension between safety and competition
The report emphasizes the need to maintain mechanisms to slow down AI development if necessary, while acknowledging that unilateral restraint could cede advantage to competitors. This reflects broader industry concerns about balancing safety with strategic positioning.
Context of massive valuation
These revelations come shortly after Anthropic announced plans for a public offering reportedly valuing the company at nearly $1 trillion, raising questions about how commercial incentives intersect with calls for caution in AI development.
CONCLUSION
The rapid shift toward AI systems that design and improve themselves signals a profound تحول in technology development, combining unprecedented acceleration with unresolved questions about control, safety, and intent.