Skip to content

AlphaEvolve leaves preview as Google sells algorithm search as a cloud tool

Original: Solve harder problems with AlphaEvolve, now available to everyone on Google Cloud View original →

Read in other languages: 한국어日本語
AI Jul 10, 2026 By Insights AI 2 min read 1 views Source

The practical question for agentic AI is moving from “can it write code?” to “can it discover better algorithms under real constraints?” Google Cloud’s general availability release of AlphaEvolve puts that question into a commercial product, not just a research showcase.

AlphaEvolve is a code optimization and discovery agent built on Gemini. Teams provide a baseline seed algorithm, a problem definition, relevant background knowledge, and a scoring function. The system then generates candidate programs, measures them against the objective, and iterates toward code that can be deployed into production workloads.

The target domains are places where brute-force human exploration runs out of road: logistics, semiconductors, genomics, high performance computing, and financial services. These are not simple autocomplete problems. They require measurable objectives, operational constraints, and enough domain structure for a candidate solution to be judged objectively.

The strongest part of Google’s launch is the customer evidence. BASF used AlphaEvolve to build a supply-chain digital twin and, according to Google, improved existing planning and forecasting models by more than 80%. Coolblue applied it to a 28-day demand forecasting pipeline and reduced WMAPE by over 5% after a few hundred iterations. FM Logistic reported a 10.4% routing improvement on top of an already optimized warehouse baseline, translating into more than 15,000 km of staff travel saved.

The release also clarifies where humans remain central. Engineers still own the benchmark, constraints, review, and release decision. AlphaEvolve narrows the search space and proposes candidates that would be too time-consuming to explore manually. That is a more grounded framing than treating agents as autonomous software departments.

The broader signal is that domain-specific evaluation may become the real enterprise AI moat. A model that scores well on generic coding tests is useful, but an agent that can lower logistics cost, chip-design friction, or HPC runtime inside a company’s own constraints is much easier to value. AlphaEvolve’s GA makes algorithm search one of the more concrete agent products now available on a major cloud platform.

Share: Long

Related Articles