The GPU has become the foundation of modern compute. Programming it - efficiently, portably, and at scale - has emerged as the most critical code generation challenge of our time. At Mako, we are building AI systems to solve this.
The Mako Kernel Agent uses cutting-edge, LLM-driven code synthesis to generate performant GPU kernels in CUDA, HIP, Triton, and beyond.
The Mako Optimization Platform integrates these kernels directly into real workloads - automatically benchmarking, tuning, and deploying them for maximum impact.
There simply aren’t enough kernel engineers on Earth to keep pace with what’s coming. Every new model architecture, every quantization technique, every novel hardware target demands bespoke GPU kernels. Today, this bottleneck forces us to build with the primitives we already have, not the ones we need.
This is one of the most severe constraints on AI innovation. The breakthroughs from frontier labs, the advances in sparsity, the pioneering work in modalities like video, DNA sequencing, and more - all of them relied on hand-tuned GPU kernels to function.
If we can automate the generation of high-performance kernels, we won’t just accelerate AI. We’ll unlock entirely new categories of algorithms and models. Mako’s mission is to tear down this bottleneck and give the world a system that speaks GPU natively, at the speed of research. If you're excited about building the foundation for advanced coding agents and solving the hardest challenges in software engineering, join us.
Send us a message and we'll get back to you.