Exciting to see how fast agentic AI chip design is making its way into academia. This semester, ChipAgents’ multi-agent root cause analysis system has been incorporated into coursework at Harvard University bringing real-world AI-driven chip design insights directly into the classroom. Today, our team also had the opportunity to give a guest lecture on superhuman chip design at Stanford University EE sharing an industry perspective on multi-agent AI for chip design and verification. One analogy that resonated: the relationship between foundation models and agentic AI is like an F1 team. McLaren can beat Mercedes in 2024 and 2025 while still using Mercedes engines. The LLM is the engine, but building a winning F1 car is far more complex. And even then, teaching someone to drive that car at the highest level is another challenge entirely. That’s where agentic systems come in. The intersection of academia and industry is where real breakthroughs happen, and it’s energizing to engage with the next generation of engineers on these topics. Thanks Stanford!

Excited to share that I’ll be speaking at Stanford next Tuesday with Ivan Bercovich (GP at ScOp VC) for a fireside chat hosted by BASES. We’ll dive into the agentic revolution at the intersection of AI and hardware, and how it’s reshaping the future of semiconductor design and manufacturing. We’ll also cover: • The current state of AI and where it’s heading • Why AI is fundamentally changing the semiconductor industry • My journey from academia to building ChipAgents 📅 April 7 🕕 6:15–7:15 PM 📍 Stanford University If you’re interested in AI, chips, or the future of intelligent systems, would love to see you there.

🚀 ChipAgents is the coolest agentic AI platform company in town! What an incredible week at DVCon. Seeing our booth and talks packed with engineers, verification leaders, and chip designers from across the industry reminds us why we started ChipAgents in the first place: to bring AI agents into the heart of semiconductor engineering. Over the past year, what began as an idea has turned into a movement. Today, ChipAgents is helping design and verification teams rethink how chips are built, from UVM testbench generation to root-cause analysis and multi-agent AI collaboration. But the most inspiring part of this journey isn’t the technology. It’s the people. We are just getting started. The future of chip design will be built by AI-native engineering teams, and we’re proud to be building that future together. Let’s go ChipAgents. 🔥

We’re excited to share that Samsung Electronics' venture arm has made a strategic investment in ChipAgents. This partnership reflects a shared conviction: as chips become more complex and design cycles tighten, AI agents will be foundational to how silicon is designed, verified, and brought to production. Samsung is one of the world’s most advanced semiconductor innovators, spanning memory, logic, and foundry. Their decision to invest underscores the real, production-level impact ChipAgents is already delivering across RTL development, verification, and engineering productivity for leading chip teams. We’re grateful for Samsung’s trust and excited to work together on pushing the frontier of AI-native chip development. The future of semiconductors will be built by humans and AI agents, side by side.

#NeurIPS2025 with the ChipAgents crew! 🚀 Pushing boundaries on AI for chip design and verification, one breakthrough at a time. Come find us and feel free to talk to our guys: we’d love to chat. 🤝

Congratulations Mehir Arora and ChipAgents team for winning the Best Tutorial Award on “How Agentic AI is reinventing chip design and verification” at DVCon Japan 2025!

DeepPath hits 1,000 citations! 🎉 Back in 2017, we introduced DeepPath, one of the earliest reinforcement learning frameworks for multi-hop reasoning over discrete knowledge graph structures. The idea was simple but ambitious: use an RL agent to learn relational paths in large-scale knowledge graphs—not by memorizing facts, but by reasoning over them. When I first floated the idea to some senior PhD students, many said it couldn’t be done—the search space was just too large. But @XiongWenhan came up with a clever supervised RL warm-up technique that tamed the RL model and made the entire system trainable. Fast forward to today, and it’s humbling to see this work cited 1,000 times. DeepPath paved the way for many of our later ideas on RL-based reasoning and inspired our current work on AI agents that operate in symbolic environments. Huge shout-out to my incredible students: Wenhan Xiong, now training Grok at xAI Thien Hoang, continuing to innovate at Google Even though they’ve long since graduated, their early contributions continue to ripple through the field. Thank you to the research community for building on this work. Here’s to the next chapter in agentic AI! 🧠🤖