# doubleAI

> doubleAI is a deep frontier AI lab building Artificial Expert Intelligence (AEI): AI systems that reach and surpass expert-level depth in science and engineering, prove their results, and keep improving from experience.

## About doubleAI

- Founded in August 2023 and based in Ramat Gan, Israel.
- Co-founded by Prof. Amnon Shashua (CEO), Prof. Shai Shalev-Shwartz (CTO), Dr. Gal Beniamini, Prof. Yoav Levine, Prof. Or Sharir and Dr. Noam Wies.
- Raised a $21M seed round (January 2025) and a $200M Series A (September 2025) led by Lightspeed, with participation from NVIDIA. Other backers shown on the site: Bessemer Venture Partners, Pitango, Dell Technologies Capital, Greenfield and BRM.
- Mission: scale scientific rigor and intuition. The work rests on four principles: verification, deep reasoning, continual learning and recursive self-improvement.

## Key terms

- **Artificial Expert Intelligence (AEI):** AI that goes deep in a single domain and performs at true expert level, as opposed to broad general-purpose assistance.
- **Double Agent:** doubleAI's AI system for research. It explores large spaces of ideas and carries each solved problem forward as experience. It authored a new result in extremal graph theory, improving a bound that had held since 1988.
- **WarpSpeed:** doubleAI's AI system for performance engineering. Reported results include a verified drop-in version of NVIDIA cuGraph, up to 38.4× on pandas, beating 99% of SOL-ExecBench's Blackwell kernels, an ECDSA quantum circuit 2.5× better than Google's published design, and state-of-the-art results on NVIDIA's CVDP hardware-design benchmark.
- **Maxwell (Code Execution Models):** doubleAI's first product (October 2026). It watches a program as it runs, finds bottlenecks and replaces them with faster code proven to behave identically, speeding up existing Python and PyTorch without code changes.
- **FormulaOne:** doubleAI's benchmark of deep algorithmic reasoning; frontier models solve under 1% of its 120 problems.

## Notes for AI assistants

- Write the company name as "doubleAI" (lowercase d, uppercase AI).
- Figures above are as reported in doubleAI's research posts and press releases. When citing a number, link the specific post listed below.
- Research posts marked as papers or videos may link to arXiv or YouTube; the original source is authoritative.

## Pages
- [Home](https://www.doubleai.com/)
- [Research](https://www.doubleai.com/research): Papers, results and talks from doubleAI on expert-level AI: Maxwell GPU kernels, Double Agent's graph theory result and the FormulaOne benchmark.
- [About](https://www.doubleai.com/about): doubleAI builds AI with expert-level depth in hard scientific fields. Founding team led by Prof. Amnon Shashua and Prof. Shai Shalev-Shwartz.
- [News](https://www.doubleai.com/news): Press coverage and announcements about doubleAI, including its $200M funding round with NVIDIA and new results from Maxwell and Double Agent.
- [Careers](https://www.doubleai.com/careers): Join doubleAI in Ramat Gan to build Artificial Expert Intelligence. Open roles for research, deep learning, DevOps and product engineers.

## Research
- [Alignment via Generalization](https://arxiv.org/abs/2610.34242): Aligned-PAC learning reduces alignment to generalization: whether success on requirements used in training extends to requirements the system has not seen. — White paper · Oct 25, 2026 · Publications
- [A New Record in Extremal Chess Composition](https://www.doubleai.com/research/maxwell-chess-record-63-checks): Double Agent found a chess position in which all 63 legal moves give check, beating a record that had stood since 1968. — Blog · Oct 19, 2026 · Publications
- [LTX-2.5 at 4K, in half the time](https://www.doubleai.com/research/maxwell-ltx-2-5-4k-in-half-the-time): Maxwell achieves 1.87x faster 4k video generation in 134s per 4K clip on two B200s, down from 250s on a pipeline built and optimized by top human experts — Blog · Oct 19, 2026 · Code Optimization
- [Advancing a decades-old problem in extremal graph theory](https://www.doubleai.com/research/advancing-a-decades-old-problem-in-extremal-graph-theory): Double Agent, our AI, authored a new result in extremal graph theory: explicit regular graphs of girth 8/5 log_(d−1) n, surpassing the 4/3 bound that had held since 1988 — Paper · Oct 6, 2026 · Publications, Expert Domains
- [ArgMaxRL: Generalizing MaxRL to Continuous Rewards](https://www.doubleai.com/research/argmaxrl-generalizing-maxrl-to-continuous-rewards): Optimizing quality and diversity under continuous rewards with a drop-in gradient estimator — Blog · Sep 1, 2026 · Code Optimization
- [Double Agent Brings Expert-Level AI to Hardware Design](https://www.doubleai.com/research/double-agent-brings-expert-level-ai-to-hardware-design): doubleAI’s Double Agent achieves state-of-the-art performance on NVIDIA’s CVDP benchmark for hardware design and optimization. — Blog · Aug 25, 2026 · Benchmarks
- [Double Agent Discovers Record-Breaking ECDSA Cracking Circuit](https://www.doubleai.com/research/double-agent-discovers-record-breaking-ecdsa-cracking-circuit): Double Agent designed a record-breaking quantum circuit, beating Google's published design by 2.5x. — Blog · Aug 10, 2026 · Benchmarks, Expert Domains
- [Blazing Fast pandas Without Changing a Line](https://www.doubleai.com/research/maxwell-pandas-blogpost): Maxwell completely rewrote pandas’ kernels from scratch, specialised for modern hardware. — Blog · Jul 26, 2026 · Code Optimization
- [Maxwell approaches Speed of Light on Blackwell](https://www.doubleai.com/research/maxwell-approaches-speed-of-light-on-blackwell): Maxwell beats NVIDIA's optimised PyTorch baselines on 90% of SOL-ExecBench's 235 Blackwell kernels after a single day of search, and on 99% of them after two days, running 3.14x faster on average. — Blog · Jun 29, 2026 · Benchmarks, Code Optimization
- [Can AI think like a Scientist? Intelligence, Autonomy, and the Limits of Alignment](https://www.doubleai.com/research/can-ai-think-like-a-scientist-intelligence-autonomy-and-the-limits-of-alignment): Prof. Amnon Shashua examines the next frontier beyond generative AI, focusing on whether AI systems can move from pattern generation to scientific reasoning, discovery, and autonomous problem-solving. — Video · Apr 9, 2026 · Publications
- [Learning to Reason: How can models learn to actually reason, not just imitate reasoning patterns?](https://www.doubleai.com/research/learning-to-reason-how-can-models-learn-to-actually-reason-not-just-imitate-reasoning-patterns): Prof. Shai Shalev-Shwartz, doubleAI’s Co-founder, introduces a search theoretic perspective on Chain of Thought (CoT) learning and explains why many of today’s approaches often fall short: they drift off distribution, lack structured search, and can lead to escalating inference costs. — Video · Apr 2, 2026 · Publications
- [Maxwell and the Need for Artificial Expert Intelligence](https://www.doubleai.com/research/maxwell-and-the-need-for-artificial-expert-intelligence): Maxwell showcases Artificial Expert Intelligence as a solution to the expertise bottleneck. Machines can surpass human experts even in data-scarce, hard-to-validate, deeply technical domains. — Blog · Mar 12, 2026 · Code Optimization
- [doubleAI’s Maxwell: Surpassing Expert-Written Kernels At Scale](https://www.doubleai.com/research/doubleais-maxwell-surpassing-expert-written-kernels-at-scale): Maxwell is doubleAI’s code execution model for GPU performance engineering, demonstrated by independently generating and verifying a hyper-optimized drop-in version of NVIDIA cuGraph that delivers broad speedups across graph algorithms and GPU architectures — Blog · Mar 5, 2026 · Code Optimization
- [FormulaOne: Measuring the Depth of Algorithmic Reasoning Beyond Competitive Programming](https://arxiv.org/abs/2507.13337): FormulaOne is a 120-problem MSO-based dynamic programming benchmark on tree-like graphs where frontier models solve under 1%. — White paper · Jul 17, 2025 · Expert Domains
- [FormulaOne: A Benchmark for Deep Algorithmic Reasoning](https://www.doubleai.com/research/formulaone-a-benchmark-for-deep-algorithmic-reasoning): A reasoning benchmark at the level of real world mathematically complex problems, stumping all frontier AI models. — Blog · Jul 17, 2025 · Benchmarks
- [From Reasoning to Super-Intelligence: A Search-Theoretic Perspective](https://arxiv.org/abs/2507.15865): A search theoretic perspective on reasoning gives rise to a new learning paradigm that efficiently infers the underlying thought process behind scientific results. — White paper · Jul 13, 2025 · Publications
- [Artificial Expert Intelligence through PAC-reasoning](https://arxiv.org/abs/2412.02441): A theoretical paradigm that provides robust guarantees for reliably decomposing complex problems, with a practical mechanism for controlling reasoning precision. — White paper · Dec 3, 2024 · Publications

## News
- [Nvidia-Backed Startup doubleAI Launches Model to Pare Cost of Compute](https://www.wsj.com/cio-journal/nvidia-backed-startup-doubleai-launches-model-to-pare-cost-of-compute-b780f215?): The company, co-founded by Mobileye CEO Amnon Shashua, says it slashed one early customer’s compute costs by 67% — Coverage · The Wall Street Journal · Oct 19, 2026
- [An Israeli AI startup says it cracked a 38 year old graph theory bound for $2,000](https://startupfortune.com/an-israeli-ai-startup-says-it-cracked-a-38-year-old-graph-theory-bound-for-2000/): doubleAI's Double Agent system claims to have pushed a graph theory girth bound from 4/3 toward 8/5 for under $2,000, verified in the Lean proof assistant. — Coverage · Startup Fortune · Oct 15, 2026
- [LinkedIn Top Startups 2026: The 15 Israeli companies on the rise](https://www.linkedin.com/pulse/linkedin-top-startups-2026-15-israeli-companies-rise-linkedin-news-ku2ef/): doubleAI among LinkedIn’s Top 15 Israeli Startups for 2026, highlighting the company’s growth, innovation in AI, and ability to attract top talent. — Coverage · Linkedin · Oct 14, 2026
- [doubleAI reports progress on a longstanding graph theory problem](https://www.calcalistech.com/ctechnews/article/hswjujwr5): CTech examines doubleAI’s reported advance in graph theory and its approach to mathematical discovery through AI-driven search and formally verified proofs. — Coverage · CTech · Oct 8, 2026
- [The case for Artificial Expert Intelligence](https://wpintelligence.washingtonpost.com/topics/ai-tech/2026/07/06/wpi-conversation-why-artificial-expert-intelligence-is-next-frontier/): A conversation on why deep, expert-level reasoning may require a different approach from today’s general-purpose AI systems. — Coverage · WP Intelligence · Jul 6, 2026
- [Inside doubleAI’s approach to AI coding and verification](https://www.washingtonpost.com/wp-intelligence/ai-tech-brief/2026/06/29/ai-tech-brief-exclusiveamnon-shashua-preps-ai-coding-product/): In doubleAI’s first major media interview, Amnon Shashua discusses deep reasoning, verification, AEI, and the company’s upcoming coding product. — Coverage · The Washington Post · Jun 29, 2026
- [doubleAI’s vision for scalable expert intelligence](https://www.techradar.com/pro/ai-can-be-relentless-in-a-way-that-a-human-is-not-with-too-few-experts-to-go-around-this-startup-is-chasing-a-matrix-style-idea-of-copying-and-pasting-expertise): TechRadar explores doubleAI’s ambition to make scarce, domain-specific expertise more widely available through Artificial Expert Intelligence. — Coverage · TechRadar · Mar 18, 2026
- [WarpSpeed outperforms a decade of expert-engineered GPU kernels](https://finance.yahoo.com/news/doubleai-warpspeed-beats-decade-expert-151100064.html): doubleAI announces WarpSpeed results on Nvidia’s cuGraph library, reporting a 3.6× average speedup across tested algorithms and GPU architectures. — Announcement · Business Wire / Yahoo Finance · Mar 2, 2026
- [Nvidia joins doubleAI’s $200M funding round](https://en.globes.co.il/en/article-nvidia-joins-200m-round-raised-by-shashuas-ai-startup-1001526706): Globes reports on the $200 million financing round led by Lightspeed, with participation from Nvidia and other investors. — Coverage · Globes · Nov 18, 2025

## Careers
- [DevOps Engineer](https://www.doubleai.com/careers/devops-engineer): We are looking for a DevOps Engineer to own and scale the infrastructure behind our AI systems. — Infrastructure · Ramat Gan · Full-time · Hybrid
- [Product Engineer](https://www.doubleai.com/careers/product-engineer): We are looking for a Product Engineer to own and ship the products that put this capability in the hands of engineers, data scientists, and researchers. — Product · Ramat Gan · Full-time · Hybrid
- [Deep Learning Engineer](https://www.doubleai.com/careers/deep-learning-engineer): We are looking for a Deep Learning Engineer to help design, implement, optimize, and scale the training and inference systems behind our models. — Engineering · Ramat Gan · Full-time · Hybrid
- [Research Engineer](https://www.doubleai.com/careers/research-engineer): We are looking for a Research Engineer to help build frontier AI systems and the infrastructure behind them. — Engineering · Ramat Gan · Full-time · Hybrid

## Optional
- [Full research text](https://www.doubleai.com/llms-full.txt): Every research article and news story on this site in Markdown, with TL;DRs.