AI Starts to Self-Code and Build Themselves

2026-06-26 · Source: AI Advances - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Expert, quick

Summary

In May 2025, Google DeepMind's AlphaEvolve system achieved a significant breakthrough by discovering a new algorithm for multiplying 4x4 complex-valued matrices, using only 48 scalar multiplications. This improved upon Volker Strassen's 1969 record of 49, a mathematical benchmark that had stood for 56 years and resisted previous attempts, including by DeepMind's AlphaTensor. Crucially, AlphaEvolve found this algorithm without human intervention. The system then applied this self-improvement capability to its own training infrastructure, making it faster. This development signifies the emergence of an AI capable of enhancing the very process that trains future AI, initiating a recursive self-improvement loop.

Key takeaway

For AI Scientists and Research Scientists designing next-generation systems, recognize that AI is evolving beyond static tools. You should prioritize research into recursive self-improvement mechanisms, as demonstrated by AlphaEvolve's ability to autonomously discover algorithms and optimize its own infrastructure. This shift implies a need to consider AI systems that can independently enhance their core capabilities, fundamentally altering development paradigms and accelerating progress.

Key insights

AlphaEvolve demonstrates AI's capacity for autonomous algorithmic discovery and recursive self-improvement, potentially leading to qualitatively new systems.

Principles

• AI can autonomously discover superior algorithms.
• Recursive self-improvement creates a feedback loop.
• AI can optimize its own training infrastructure.

In practice

• Explore AI systems for self-optimization tasks.
• Investigate AI-driven algorithmic discovery.

Topics

• AlphaEvolve
• Algorithmic Discovery
• Recursive Self-Improvement
• AI Optimization
• Matrix Multiplication
• DeepMind

Best for: AI Scientist, Research Scientist, Director of AI/ML

Related on AIssential

• Import AI 455: Automating AI Research — AI systems are rapidly automating core R&D tasks, making autonomous AI development probable by 2028.
• this EX-OPENAI RESEARCHER just released it... — AI agents can autonomously conduct machine learning research, improving model training and performance.
• Recursive Superintelligence emerges from stealth with $650M raise — Recursive self-improvement is posited as the fastest path to achieving superintelligence.
• GPT-5.5: The paper is important because it shows an early version of AI becoming part of the machinery that improves AI. It does not prove that fully autonomous science has arrived. — AI systems can now automate and accelerate core aspects of AI research and development.
• Hard Takeoff has started — AI models are now recursively self-improving, accelerating development and reducing human bottlenecks.
• AlphaEvolve: How our Gemini-powered coding agent is scaling impact across fields — AlphaEvolve, a Gemini-powered agent, autonomously discovers and optimizes algorithms, driving significant advancements across science, infrastructure, and commercial applications.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Advances - Medium.