Agentrys is an applied research team building the future of chip design. We’re developing Agentic Design Automation (ADA) — self-improving agents with an agent-native toolchain — to accelerate chip design toward full autonomy.
Backed by leading strategic investors and industry angels.
The Challenge
Chip design is not one problem — it is hundreds of interdependent tasks, many of which require optimizing performance objectives under strict constraints. While EDA tools have automated a large portion of the design process, significant manual effort remains. Design entry (RTL, schematic), verification, and analog layout are still largely manual. And even on tasks that EDA tools handle well — such as physical design — engineers still spend enormous time on analysis, optimization, and debug.
The interdependency, complexity, and multi-constraint nature of these tasks make it exceptionally difficult to find good solutions. EDA tools do their best to optimize, but long feedback loops, vast search spaces, and limited compute mean that optimal design points remain out of reach.
There is a deeper problem too. Doing this work well requires judgment that only accumulates over years of practice — and that judgment walks out the door when senior engineers retire. Meanwhile, the demand for AI chips continues to surge, outpacing the supply of qualified chip design talent.
Can AI solve these challenges and realize the dream of fully autonomous design — spanning hardware-software co-design, architecture, front-end implementation, verification, physical design, analog, and package co-design through sign-off — pursuing optimization autonomously and resolving bugs without hand-holding?
The Solution — Agentic Design Automation (ADA)
The advancement of generative AI makes this possible. Many manual design tasks — RTL entry, testbench generation, and more — can now be automated through natural language interaction with LLMs. Agentic systems go further: they actively pursue goals by running tools, interpreting results, deciding what to do next, adjusting their approach, and iterating until the objective is met. This dramatically reduces the manual burden of analysis, optimization, and debug across design tasks.
But generative AI alone is not enough. Models hallucinate and cannot directly solve high-dimensional optimization and analysis problems. The path forward is Agentic Design Automation: AI agents working in concert with EDA tools and agent-native tools — purpose-built for agents — to autonomously design chips.
Traditional EDA tools were built for human users. They are monolithic, integrated into fixed design flows, and optimized for human interaction patterns. Agents are fundamentally different: scalable, capable of making split-second decisions, and able to interact with tools at high frequency. Agent-native tools — tools built from the ground up for agents — are essential to fully realizing this potential. These tools will initially be built by humans, but increasingly, agents will build them too.
To further enhance optimization capability, we are developing GPU-accelerated engines and AI-driven methods — including reinforcement learning, evolutionary search, and learned surrogates — all natively integrated into the ADA framework. Together, these enable agents to automate chip design at a scale and quality no human team can match. The result is not just faster execution of existing tasks, but measurably better design outcomes.
Self-Improving: A System That Compounds
What sets Agentic Design Automation apart from traditional EDA tools is that it gets better over time. Agents and agent-native tools improve through three compounding loops: they learn from existing design data, they are continuously refined through expert feedback — encoding the hard-won judgment that no dataset alone can capture — and they learn from themselves, with every design run generating new training signal that improves both how agents reason and how the underlying tools evolve.
The result is a system that compounds at every level: more capable agents, better tools, and deeper design intelligence with every design completed, every expert interaction, and every closed loop.
Your team’s institutional knowledge gets encoded and amplified every time you use it.
Team
Agentrys is built by researchers and engineers who have helped define modern AI for chip design and EDA — creating pioneer systems and benchmarks — and delivering production-grade EDA, LLM training, and silicon infrastructure across NVIDIA, Meta FAIR, AMD/Xilinx, Samsung, Oracle, and IBM.
Our team combines frontier agent research with deep end-to-end chip expertise spanning hardware-software co-design, RTL and verification, PPA optimization, physical design engines, DTCO, and large-scale design infrastructure — backed by flagship EDA best paper awards, major research honors, and experience shipping real systems at scale.
News
February 2026
State-of-the-art results across four CVDP RTL coding tasks
Agentrys achieves 95.8% overall pass rate on four RTL coding tasks of the Comprehensive Verilog Design Problems benchmark, including a perfect 100% on Code Debug, through seven generations of self-improving agents.
Work with us
We're working with design teams, EDA practitioners, and semiconductor organizations who are ready to build differently. If that's you, we'd like to talk.
Get in touch: info@agentrys.ai