doubleAI today introduced a new category of AI model: Code Execution Models (CEMs). CEMs place intelligence somewhere it has never been before: between code and silicon. While coding agents act on behalf of the user, a CEM acts on behalf of the hardware. The promise of CEMs is in multiplying available hardware, through their ability to unlock a vast share of compute that sits idle today. The first CEM, doubleAI’s Maxwell, is now open for early access at heymaxwell.ai.
A CEM lives inside the running program. Unlike coding agents, which take human instructions and produce source code, CEMs operate on a different level entirely. Their inputs are the running program itself, the hardware it runs on, and the data flowing through it, and they produce instructions meant directly, and only, for the machine. The source code people and coding agents write and read stays exactly as it was. Maxwell is a CEM that continuously inspects and improves the binary under the code as it runs, fitting it to the exact hardware, data, and usage it encounters.
Every improvement Maxwell makes is formally proven correct. Until now, AI has never been trusted inside running software: changes made while code runs get no human review, and conventional AI models can make mistakes. doubleAI has made breakthroughs in verification that change this. Maxwell formally and mathematically proves that every optimization it unlocks leaves the program’s behavior unchanged.
Compute has become one of the world’s scarcest resources. CEMs get more out of every piece of hardware, either CPU or GPU. The same hardware does more work, or the same work needs less hardware. On real-world workloads, doubleAI’s first CEM, Maxwell, has made data pipelines over 38x faster and AI inference kernels over 4.5x faster.
Why Code Execution Models?
CEMs change the paradigm of how we use computers. Up until now, software has been static. Humans and AI agents write code, and that code is then carried out by the machine, exactly as written. But the environments it runs in are dynamic. Hardware, data, and usage shift constantly, and the best version of a program can change by the hour. Static code can’t keep up. This is why hardware is so underutilized.
CEMs make code dynamic. A living thing that shifts and improves as it runs. In a compute-constrained world, that is what it takes to get full use out of every chip. A CEM is an agent for the machine: it makes the CPUs and GPUs themselves intelligent.
“AI has changed how software is written, but not how it runs,” said Prof. Amnon Shashua, co-founder and CEO of doubleAI. “Code Execution Models change that. 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.”
How Maxwell Works
Maxwell watches code as it runs, rewrites the binary under the code, and fits it to exactly what the machine is doing.
- Proven correct: Maxwell formally proves every rewrite is correct, with mathematical certainty.
- Source code never changes: Only the machine instructions underneath evolve during runtime. Coding agents can still write the programs. Maxwell makes sure they run optimally.
- CPUs and GPUs: Maxwell runs on both, optimizing complex workloads.
- Self-improving code: Every proven, faster version becomes the starting point for the next, so programs keep getting faster for as long as they run.
“Much of the world’s compute goes to running code far more slowly than it could,” added Shashua. “Every program Maxwell speeds up gives some of that back.”
About doubleAI
doubleAI is a frontier lab whose team of elite Ph.D. researchers spans computer science, mathematics, physics, and biology. Progress in science and engineering has always been bounded by human expertise. doubleAI is building AI that out-reasons experts. doubleAI created Code Execution Models (CEMs), a new category of AI model that works inside running software and proves every change correct. doubleAI raised $200M in a Series A financing round co-led by Lightspeed Ventures and Bessemer Venture Partners. The investor group also includes Dell Technologies Capital, Greenfield, NVIDIA, Pitango, and BRM.
View source version on businesswire.com: https://www.businesswire.com/news/home/20261019689807/en/
Media gallery
