HyperBench, an agentic engineering platform for hardware and software co-development, has officially launched. Founders Justin Leong and Ivan G. said they saw how separate hardware and software workflows made integration failures difficult to trace. They are building HyperBench to bring system context, agent development and hardware testing into a single environment.
The Y Combinator-backed startup says coding agents often lack the context needed to understand how a software change affects a physical system. HyperBench maps repositories, hardware components, interfaces, protocols and test evidence into a unified system graph, giving AI agents the context to understand how different components interact.
Agents use this graph to trace failures across components, delegate fixes across repositories and test changes on real hardware through hardware-in-the-loop testing.
READ: AI shopping agents threaten Amazon and Walmart’s lucrative advertising businesses (November 5, 2026)
In a system-level reasoning benchmark, HyperBench reported 20% higher correctness and faster debugging compared with vanilla Claude and Codex.
According to the company’s website, the workflow involves four steps. First, the platform traces issues across code and hardware. It then captures the failure on the user’s test bench, delegates fixes across repositories and retests the changes on hardware before presenting draft pull requests for review.
The startup says its platform allows users to retain control over their infrastructure through local execution and team-based system sharing. Users can also connect their existing hardware test benches rather than investing in new equipment.
READ: Walmart lets AI agents shop for customers as Amazon cracks down on rivals (October 22, 2026)
The platform provides evidence at every handoff, allowing users to review code, test results and recordings together.
HyperBench says it was built by engineers with experience at OpenAI and Anduril. Its founders, Stanford-trained computer science and physics engineers, identified the challenges created by disconnected hardware and software workflows and developed the platform to address them.
Leong previously worked in software engineering at Anduril, where he focused on Roadrunner guidance, navigation and control (GNC) firmware. Ivan G. previously served as a member of technical staff at OpenAI, working on software infrastructure.