Ex-OpenAI and Anduril engineers build AI agents for hardware-software development

# HyperBench Launches AI Platform to Connect Hardware and Software Development

HyperBench, a new agentic engineering platform designed to bring hardware and software development together, has officially launched. Founded by Justin Leong and Ivan G., the startup aims to address a major challenge in engineering: identifying and resolving integration problems that occur when hardware and software development operate in separate workflows.

The founders said they recognised that disconnected development environments often make it difficult for engineers to determine why a software change causes problems in a physical system. HyperBench was created to bring system context, AI agent development and hardware testing into one integrated environment.

Backed by Y Combinator, the company argues that many coding agents lack the technical context required to understand how software modifications affect physical hardware. To address this limitation, HyperBench creates a unified system graph that connects code repositories, hardware components, interfaces, communication protocols and testing evidence.

This shared system map gives AI agents a broader understanding of how different parts of an engineering system interact. Instead of examining software changes in isolation, agents can use the connected information to investigate problems across multiple components and identify the source of failures.

The platform allows agents to trace issues across hardware and software, assign repair tasks across different code repositories and test proposed modifications on physical equipment. Its hardware-in-the-loop testing capability enables changes to be evaluated on actual hardware rather than relying entirely on software simulations.

HyperBench also reported improvements in its internal system-level reasoning benchmark. According to the startup, its platform achieved 20% higher correctness and faster debugging compared with standard versions of Claude and Codex.

The company describes its workflow as a four-stage process. Initially, the platform investigates problems by tracing them across software code and hardware components. It then reproduces and captures the failure using the customer's test bench before delegating potential fixes across relevant repositories.

After the proposed changes are developed, the system retests them on physical hardware to verify whether the issue has been resolved. The process concludes with draft pull requests that engineers can examine and approve before any changes are incorporated into their projects.

The startup says the platform is designed to give engineering teams control over their infrastructure. It supports local execution and team-based system sharing, allowing users to work within their existing environments. Companies can also connect their current hardware testing equipment instead of purchasing additional testing infrastructure.

Another feature is its evidence-based review process. HyperBench provides supporting information at every stage of the workflow, bringing together code modifications, test results and recordings. This allows engineers to examine what happened, understand the changes made by AI agents and evaluate the results before accepting proposed fixes.

The company says it was developed by engineers with professional experience at OpenAI and Anduril. Both founders have academic backgrounds in computer science and physics from Stanford and identified the difficulties caused by disconnected hardware and software development processes.

Justin Leong previously worked as a software engineer at Anduril, where he focused on Roadrunner guidance, navigation and control firmware. His experience involved software development connected directly to physical systems, giving him exposure to the challenges of coordinating software with hardware performance.

Co-founder Ivan G. previously worked as a member of technical staff at OpenAI, concentrating on software infrastructure. Their combined experience helped shape HyperBench's approach to bringing AI-assisted software development and physical hardware testing into a shared engineering workflow.

With the launch, HyperBench is positioning itself as a platform for engineering teams that need AI agents to do more than generate code. Its broader objective is to help developers understand system-wide problems, coordinate fixes and verify changes against real hardware while keeping engineers involved in the final review process.