doubleAI launches Maxwell: Amnon Shashua’s "code execution model" makes software faster while it runs

• Who: doubleAI, co-founded by Amnon Shashua (Mobileye founder and former CEO) and Shai Shalev-Shwartz.
• What: "Code Execution Models" (CEMs) — AI that optimizes running software; first model Maxwell, now in early access at heymaxwell.ai.
• When: announced October 19, 2026.
• Funding: $200M Series A co-led by Lightspeed Ventures and Bessemer Venture Partners, with Dell Technologies Capital, Greenfield, NVIDIA, Pitango, and BRM; valued at over $1B (per Calcalist); about 40 employees.
• Performance (company-reported, unverified): data pipelines over 38x faster, AI inference kernels over 4.5x faster; The Wall Street Journal reports one early customer cut compute costs ~67% and LTX said Maxwell doubled its model-inference speed.
• The technical core: Maxwell formally and mathematically proves every runtime optimization preserves the program’s behavior; source code never changes; runs on CPUs and GPUs; self-improving.

TEL AVIV — doubleAI, the new AI company founded by Mobileye founder Amnon Shashua, unveiled its first product on October 19, 2026: a new category of AI model it calls "Code Execution Models" (CEMs), and its first model, Maxwell, now in early access. The pitch is simple and ambitious: instead of helping humans write code, CEMs live inside running software and make it execute faster — multiplying, as the company puts it, "available hardware" out of compute that sits idle today.

Unlike coding agents, which take human instructions and produce source code, a CEM operates "between the code and the silicon." Its inputs are the running program itself, the hardware it runs on, and the data flowing through it; its output is instructions meant directly, and only, for the machine. "AI has changed how software is written, but not how it runs," said Shashua, doubleAI’s co-founder and CEO. "CEMs, and Maxwell in particular, are agents on behalf of the hardware as opposed to code agents that are on behalf of the human user. In other words, they are the first AI that works for the machine, not the programmer."

The verification breakthrough

The reason no AI has lived inside running software before, the company argues, is trust: changes made while code runs get no human review, and conventional AI models can make mistakes. doubleAI says its breakthrough is formal verification — Maxwell "formally and mathematically proves that every optimization it unlocks leaves the program’s behavior unchanged." Shashua told The Wall Street Journal that learning how to verify the code was the central challenge of making CEMs work, and that the company found a method of doing it formally and mathematically.

Maxwell continuously inspects the binary under the code as it runs and refits it to the exact hardware, data, and usage it encounters. The source code that humans and coding agents write stays exactly as it was; only the machine instructions underneath evolve — and every proven, faster version becomes the starting point for the next round of optimization, so programs keep getting faster for as long as they run. It works on both CPUs and GPUs.

38x faster pipelines — company numbers, for now

doubleAI’s reported figures: on real-world workloads, Maxwell made data pipelines over 38x faster and AI inference kernels over 4.5x faster. Reporting published October 19 by The Wall Street Journal adds that one early customer reduced computing costs by approximately 67% — a company-reported result whose applicability across other workloads has not been independently established — and that customer LTX told the Journal Maxwell doubled its model-inference speed. As Ctech noted, the figures have not been independently verified and the announcement does not provide enough detail about the underlying tests, hardware configurations, or baselines to assess how broadly the gains can be replicated.

The timing is deliberate. Citing SemiAnalysis data, the Journal reported that the cost-per-hour of renting Nvidia H100 chips on a one-year contract has climbed 60% in the last year as demand outstrips supply. "More hardware is being built, but the need for compute is rising faster than the ability to build GPUs and CPUs," Shashua told the Journal. "If you can utilize the hardware better, then you can achieve something incredible in terms of savings." doubleAI plans to charge customers based on the computing cost savings it delivers.

$200M at a $1B+ valuation, 40 people

doubleAI raised $200 million in a Series A co-led by Lightspeed Ventures and Bessemer Venture Partners, with participation from Dell Technologies Capital, Greenfield, NVIDIA, Pitango, and BRM — a round Calcalist previously reported values the company at more than $1 billion. The company, which operated in stealth until recently, employs about 40 people, according to Ctech. Shashua founded doubleAI after stepping down as Mobileye’s CEO and leaving his roles at AI21, alongside co-founder Shai Shalev-Shwartz.

Dell Technologies Capital managing partner Yair Snir said the fund had invested in many AI companies but had chosen to back a company developing its own models for the first time with doubleAI, leading its seed round. "Every few weeks, they unveil another AI breakthrough, significantly improving the results achieved by some of the best teams in the world," Snir said. "They are now applying this same mindset to the challenge of computational efficiency, using an approach that no one else has taken to date."

Whether "code execution models" become a real category or sharp positioning will turn on whether Maxwell’s verified speedups survive contact with customers’ messy production workloads. But with compute now the industry’s scarcest resource, an AI that works for the machine rather than the programmer is one of the more interesting ideas to arrive this year. Maxwell is in early access at heymaxwell.ai.

Primary source: doubleAI’s October 19 launch announcement (Business Wire).

A new AI company for the industry that prices risk

"Insurance is the quiet force that lets ambition outrun risk, and there’s an incredible opportunity to rebuild it for the millions of businesses that depend on it," said Rushil Goel, founder and CEO of Mandl, in the company’s launch announcement on Monday. "We’ve already proven that an AI-native model works across underwriting, claims, and operations to improve how businesses price risk, recover from losses, and get coverage that fits. Launching Mandl is the natural next step."

Mandl’s pitch is blunt: commercial insurance still prices most businesses by category and averages, the way it has for decades. The company plans to expand across the value chain in two ways — building new AI-native companies, as it did with Vega and Cooper, and acquiring firms that share its vision. Every Mandl company operates as an independent business, with backing to grow faster than it could alone.

The portfolio: Nirvana, Vega, and Cooper

The engine room is Nirvana, the AI-native commercial insurer Goel cofounded, which has grown to more than $300 million in run-rate premium with top-tier loss ratio performance. "We built Nirvana to understand risk better than anyone else, so every business pays for how it actually operates," said Jon Hammer, previously Nirvana’s general manager and now its CEO. He leads the company alongside cofounder and Chief Actuary Alex Carges, while Goel stays on as chairman. "The best operators should pay the least, and with Nirvana, they finally can," Hammer added.

Vega, Mandl’s AI-native claims administrator, is already delivering claims cycle times four times faster than the industry standard. "Businesses don’t buy insurance for the policy, they buy it for the moment something goes wrong," said cofounder and CEO Talha Zubair. "As part of Mandl, we can finally make that the standard across the industry." Cooper, meanwhile, brings an AI-native operations platform to agencies, brokerages, MGAs, and carriers, with what the company describes as a mission to make every insurance professional extraordinary. "Insurance professionals lose hours every day to work that AI should handle," said cofounder and CEO Akhilesh Koppineni.

Why Goel’s track record matters

Before Nirvana, Goel was an executive at Samsara (NYSE: IOT), where he led the Connected Fleets platform from the ground up to more than $500 million in annual revenue. That combination — an AI-native insurer already clearing $300 million in premium, plus an operator who has scaled industrial-scale software before — is the thesis investors bought into.

Mandl says it will share more about its growth in the coming months and is hiring across technology, finance, strategy, and operations. The open question is whether the build-and-acquire playbook can do to insurance’s middlemen and manual workflows what AI-native underwriting has already done to Nirvana’s risk book. The company points to the numbers: $300 million of premium priced the AI way, claims resolved in a fraction of the legacy time. If those scale across an acquired book of business, "rebuild the full engine" stops being a slogan.