A new framework lets AI agents learn from their own search histories instead of burning through expensive compute cycles
Google researchers have built a system that lets AI discovery agents get dramatically smarter without actually doing more work. Dream-RSI, a framework developed by teams at Google DeepMind, the University of Maryland, and the University of Virginia, cuts the number of required discovery-agent calls from 51,200 down to just 317 on a benchmark optimization task. That’s a roughly 162x reduction.
The paper, published on arXiv as arXiv:2609.14858, describes a system that essentially teaches AI agents to replay and learn from their own prior searches rather than running new ones from scratch every time.
How Dream-RSI actually works #
The framework operates on a three-stage loop, each building on the last. First, the system runs an initial phase of online exploration, generating what the researchers call “discovery trees.” These are structured records of every search path the agent took, including the dead ends and the breakthroughs.
In the second stage, those historical trails get transformed into replay simulators. Instead of needing to call the underlying coding agent again, the system reconstructs the decision landscape from data it already has.
The third stage is where the “dreaming” happens. A separate large language model evaluates and refines the exploration policies, things like branching strategies, batching rules, and when to stop searching, all offline. No new expensive agent calls required. The orchestration layer tests different approaches against the replay simulator, figures out what works best, and locks in improved policies for future runs.
The result is recursive self-improvement without updating the core model’s weights. The agent doesn’t need to be retrained.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The numbers tell the story #
On a Lasso path solver task, the SimpleTES baseline required 51,200 discovery-agent calls. Dream-RSI running on Gemini-3.1-Pro brought that down to 317. Average runtime dropped from 3,587.1 milliseconds to 2,931.0 milliseconds.
The solutions Dream-RSI produced outperformed established libraries like sklearn and glmnet across six held-out datasets. Even compared to a recursive fixed exploration approach that used 550 calls, Dream-RSI’s 317 calls represented a meaningful improvement.
On KernelBench, Dream-RSI achieved performance parity with existing methods while generating 2.43x fewer generations on VGG16 and 1.79x fewer on LayerNorm.
Why this matters beyond the benchmark scores #
Gemini-3.1-Pro and Gemini-3.7-Flash were used in the experiments without any fine-tuning or weight updates. Dream-RSI sits on top of existing models as an orchestration layer.
The research team, led by Tong Zheng and including contributors like Xidong Wu and Zheng Zhang from both Google DeepMind and the partnering universities, has set up a project site and GitHub repository. The code is reportedly being prepared for external release.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our