Prof. Amnon Shashua’s new AI startup doubleAI today unveiled its new code execution model called Maxwell and has made it available for early access to software developers.
The primary feature distinguishing standard development engines from this Israeli company’s new engine is the ability to modify live code at both the hardware and middleware layers, while the software is running. Unlike other language models, which convert developer-written commands into source code, Maxwell’s engine operates differently: the activity of the software and hardware, along with the data flowing between them, generates instructions directed straight to machine language - instructions that evolve over time. These capabilities were previously associated with other software components, such as optimizers that adjust code during runtime.
doubleAI says this approach also offloads a significant portion of the processing burden from the code itself, thereby saving on "tokens" (AI computational units) and reducing the need for expensive graphics processors (GPUs).
"AI has changed how software is written, but not how it actually functions," says Prof. Shashua, cofounder of doubleAI. "Execution models in general, and Maxwell in particular, change the landscape. They act as hardware agents, not merely as code-based agents representing the human user. In other words, this is the first form of AI that works for the machine, rather than for the programmer."
While the software code, or source code, remains unchanged, Shashua’s engine enhances hardware processing operations, continuously adapting them to the software running on top. He says the system operates on both graphics processors used for AI processing and central processing units (CPUs) utilized for a variety of tasks, including AI processing, storage, and application management. The development engine includes a component based on algorithms, distinct from standard Large Language Models (LLMs), that functions as a "verifier." This component examines the generated code and validates only those segments that are mathematically proven to be correct. This verification mechanism, integrated with Maxwell’s product development framework, is a core feature of doubleAI. The "verifier" is also utilized to provide formal mathematical proofs for problems in mathematics and computer science.
Last week, Shashua’s AI company announced that it had solved the "Longest Path Problem" - a major challenge in graph theory and one of the 100 most significant unsolved mathematical problems. The problem involves finding the longest path between two points within a graph, ensuring no node is visited more than once, measured by the number of edges. This has substantial practical applications, such as correcting computational errors in quantum computing and in the miniaturization of microchips. Shashua noted that the cost of solving the problem did not exceed $2,000.
To date, the company has raised $200 million from investors including Lightspeed, Bessemer, Dell Technologies Capital (DTC), Greenfield, Nvidia, Pitango, and the Barkat brothers’ BRM. In this funding round... Its most recent financing round was completed at a company valuation of $1 billion.
Yair Snir, managing partner at Dell Technologies Capital - Michael Dell’s investment fund - which led the company’s seed round, stated: "We have invested in many AI companies, but in this instance, for the first time, we chose to invest in a company that develops its own models, and we led its seed round. We saw in Amnon and his team an ability to approach complex problems from a different perspective and to move forward rapidly. Every few weeks, they unveil another AI breakthrough, significantly improving the results achieved by some of the best teams in the world. They are now applying this same mindset to the challenge of computing efficiency, utilizing an approach that no one else has taken to date. This is precisely the kind of innovation that Israel can and should lead."
Published by Globes, Israel business news - en.globes.co.il - on October 19, 2026.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2026.