Episode Summary
Executive Summary: Sergei Nestorenko, CEO of Quilter, explains how his startup uses reinforcement learning to automate PCB layout, a slow, error-prone, highly specialized part of hardware design. The conversation covers the pain of traditional board design, Quilter’s current limits and roadmap, the RL-based technology stack, market demand, pricing, and the labor and industry implications of automating electronics engineering.
Main Topics: Founding story and SpaceX inspiration: Sergei traces Quilter’s origin to his electronics work at SpaceX, where PCB design pain was a constant bottleneck. He says the challenge of building reliable hardware led him to pursue automation. How PCB design works today: The interview breaks down PCB creation into schematic capture, layout, and manufacturing, emphasizing that layout is still mostly manual and requires deep tribal knowledge plus repeated iterations. Quilter’s product scope and roadmap: Quilter currently automates only relatively basic boards by taking a completed schematic plus constraints and generating a manufacturable layout. The long-term goal is to eliminate layout entirely and move up to increasingly complex boards. Reinforcement learning as the core technique: Sergei argues Quilter is more like AlphaGo than a language model: it creates a 'game' of board design, optimizes against physics and manufacturability, and learns by exploration rather than imitation. Economics, customers, and labor disruption: The discussion covers PCB design labor costs, shortages of skilled board designers, which teams buy the product, and how automation may change demand for human layout work. Manufacturing, scale, and moat: Sergei explains how integration with manufacturers and customer-specific design patterns can create compounding advantages and defensibility as Quilter scales. Startup lessons and hiring: He emphasizes mission-driven recruiting, the emotional difficulty of startups, self-care, and why Benchmark’s experience and founder reputation were attractive to him.
Key Arguments: PCB layout is one of the major bottlenecks in hardware development because it is slow, manual, and requires specialized expertise that is often spread thin across engineers. Quilter’s aim is to make board layout feel like compiling code: a designer completes the schematic, specifies constraints, and the system generates the rest. Reinforcement learning is a better fit than supervised imitation because the best designs may lie outside human examples and need exploration to discover. Basic manufacturability checks are already automated, but the hard problems are physics interactions like crosstalk, speed, signal integrity, heat, and EMI. Even 'easy' boards can take a week or two by hand, so automating them can create significant productivity gains even before tackling top-end complexity. The market is not just board designers; many adjacent engineers need to design boards as part of their work and would benefit from automation. Quilter’s moat may come from compounding design data, customer-specific pattern learning, and deeper integration with manufacturers. Future superhuman board design could reduce size, cost, and latency while improving first-pass correctness and manufacturing yield. The industry has a shortage of PCB and electrical engineering talent as more young engineers choose software/AI/robotics instead of hardware. For many applications, the most valuable improvement is getting the board right on the first try, since debugging hardware mistakes is far harder than debugging software.
Data Points: Seed/Series funding: $10 million - Sergei says Quilter raised a $10M round led by Benchmark. PCB layout professional salary: $100,000 to $110,000 per year - Estimated median salary for specialists doing PCB layout. Annual labor spend on PCB layout: $20 billion to $50 billion per year - Sergei estimates total labor spent globally on this task. Layout time for simple boards: a couple of hours - For very simple circuit boards with few components. Layout time for complex boards: 2 to 3 months - For boards like a Starlink antenna or iPhone motherboard. Prototype turnaround time after design: 24 hours to a few days - Manufacturing can be fast if the team is willing to pay for expedited fabrication/shipping. Quilter current performance gap: not yet as good as the average designer on physics/density-related boards - Sergei says Quilter is mainly better than humans at speed today. Quilter speed advantage: about 1 hour versus 1 week - For boards where physics constraints are not too significant. Near-term roadmap: within 5 years - Sergei believes Quilter could become as good as or better than most humans at physics-heavy design within this timeframe. Chip fab throughput: 8,000 unique designs per day - He cites the biggest fab in China as producing this volume. RL training demo: 5 minutes on a single GPU - Example of training a robot to walk in simulation.
Pivotal Quotes: "We want to get to the point where doing a PCB layout is like compiling code, where, you know, in practice, pretty much in all cases, people just click a button and it runs." — Sergei Nestorenko: Explaining Quilter’s long-term product vision. "The best strategies came from the network trying new things and seeing how they work and remembering them." — Sergei Nestorenko: Arguing for reinforcement learning over imitation learning. "I want to be able to take this video call on a quilter-designed computer by the end of this year." — Sergei Nestorenko: Describing an ambitious internal milestone for the company.
Implications: If Quilter scales, PCB design could shift from artisanal, iterative labor to software-like automation, accelerating hardware development and potentially reshaping engineering roles. The biggest gains may come first in prototypes and adjacent engineering workflows, then later in mission-critical complex systems.
From the Episode
What is the designer's life like with Quilter? How much of that do you guys automate today? And how do you think that's going to evolve over the next five years? Yeah, so the principle of what we're trying to do is to completely eliminate layout altogether. We want to get to the point where doing a PCB layout is like compiling code, where, you know, in practice, pretty much in all cases, people just click a button and it runs. Now, of course, that's much easier said than done. And so that's an easy To say, but a very, very difficult thing to do. So, with Quilter, for the boards that are kind of in our scope today, which I could talk about in a second, that is very much the process, right? The designer will finish their schematic. They will open up the beginnings of their board file in order to enter some constraints. So, for example, you might have very physical requirements on how big the board can be, where connectors go, where buttons go, where a screen might go. And then everything else that you need to do, you simply leave off the board in the design file.
Actually, just difficult to get to really good designs. There's not a GitHub of millions and millions of designs out there. And furthermore, the best designs are locked away at NVIDIA, at Apple, at SpaceX, and not in open source. But second of all, I actually think long-term that's the wrong approach. I think that if you focus on learning from human examples, you're likely to peak at roughly human performance, which is actually what happened to DeepMind when they first started playing Go. They looked at a whole bunch of human examples of moves. Trained on them, queried the net to make the best human-like move, and they got to like a grandmaster level, but not a world champion level. The best strategies came from the network trying new things and seeing how they work and remembering them. And I think that's what will eventually lead to superhuman-level board designs for quilter. Yeah, it's like there's all this unexplored solution space outside of where humans have been playing. And if you can explore that, you might find much more efficient designs than what humans. Know today. Hey, we'll continue our interview in a moment after a word from our sponsors.
But we'd like to do it. I want to be able to take this video call on a quilter-designed computer by the end of this year. And I think, like, that's awesome. Yeah. I think, should we pull that off? I mean, if not this year, we'll pull it off shortly after. But, like, imagine writing some code and then eventually having a board on your desk that then you made with algorithms that then does that. Like, that is so satisfying to think about and to do. And so, certainly, that draws people to tackling this problem. Be in the computer history museum one day for an exhibit. That's no, that's awesome, and that's an incredible, incredible goal. Um, another question on talent: you know, it seems like the sort of trajectory for a typical research scientist focused on AI or machine learning broadly, kind of went through undergrad, went through a PhD program, did a bunch of research, and then went to one of the big tech companies and continued doing research there. And a lot of that research was essentially academic.
About The Cognitive Revolution
A biweekly podcast where hosts Nathan Labenz and Erik Torenberg interview the builders on the edge of AI and explore the dramatic shift it will unlock in the coming years. The Cognitive Revolution is part of the Turpentine podcast network. To learn more: turpentine.co