Krux
May 9, 2026
DeepMind's Coding Agent Cuts Genomics Errors by 30%
Published: May 9, 2026 at 12:15 AM
Updated: May 9, 2026 at 12:15 AM
100-word summary
DeepMind's AlphaEvolve has moved from lab to real-world production, and the numbers show why companies are paying attention. The Gemini-powered coding agent slashed variant-detection errors in genomics by 30% and boosted power grid optimization success rates from 14% to over 88%. A logistics company using it saved 15,000 kilometers of annual travel through smarter routing. Klarna tapped it to speed up transformer training while improving model quality. The tool is now available through Google Cloud with both UI and API access. What makes this different: AlphaEvolve writes and optimizes algorithms autonomously, not just code. It's tackling problems where human programmers hit walls.
What happened
DeepMind's AlphaEvolve has moved from lab to real-world production, and the numbers show why companies are paying attention. The Gemini-powered coding agent slashed variant-detection errors in genomics by 30% and boosted power grid optimization success rates from 14% to over 88%. A logistics company using it saved 15,000 kilometers of annual travel through smarter routing. Klarna tapped it to speed up transformer training while improving model quality. The tool is now available through Google Cloud with both UI and API access. What makes this different: AlphaEvolve writes and optimizes algorithms autonomously, not just code.
Why it matters
It's tackling problems where human programmers hit walls.