CHAPTERS
- 0:05 – 0:31
Diode’s mission: AI that automates printed circuit board (PCB) design
Diana welcomes Diode founders Davide Asnaghi and Lenny Khazan shortly after their Series A and rapid post-YC growth. They define Diode as an AI-enabled “design shop” focused on automating PCB design end-to-end.
- •Diode positions itself as AI for circuit board (PCB) design automation
- •Framing as an AI-enabled design shop rather than a traditional EDA tool
- •Context: fast growth after YC and a newly closed Series A
- 0:31 – 0:51
Who uses Diode: from Fortune 100 to robotics-first startups
Davide describes Diode’s early traction across a broad customer set, including large enterprises and cutting-edge hardware/robotics startups. The breadth signals both high demand and applicability across industries that need custom boards.
- •Customer mix includes Fortune 100 companies and smaller startups
- •Examples: Physical Intelligence (robotics foundation models) and autonomous boat builders
- •Startups use Diode to accelerate early prototyping and board delivery
- 0:51 – 2:13
Origin story: software meets hardware pain (and mentorship that became co-founding)
Lenny recounts how, during COVID, he moved from software into hardware work and was mentored by Davide. Working together later, they repeatedly saw hardware teams suffer from poor tooling and slow feedback cycles compared to software engineering.
- •Lenny transitions from software to hardware curiosity during college
- •Davide mentors Lenny, exposing him to real-world hardware development
- •Core observation: PCB workflows lack fast iteration/feedback compared to software
- •Motivation forms: improve the state-of-the-art in hardware design tooling
- 2:13 – 2:55
The first YC idea: PCB “compiler” verification to catch mistakes
Davide explains their initial concept: use LLMs to parse documentation and identify “silly mistakes” in existing PCB designs. They applied to YC aiming to provide automated verification—essentially a compiler-like checker for electronics.
- •Inspiration from custom silicon workflows and one-shot correctness expectations
- •Belief: LLMs can catch errors if given the right representation of a design
- •Initial YC pitch: customers provide designs; Diode flags mistakes
- 2:55 – 4:01
Customer discovery reality check: “We don’t make mistakes” and why it mattered
After extensive user interviews early in the batch, the team learned verification wasn’t compelling as a standalone product. Even with sophisticated internal approaches (e.g., generating boards and injecting mistakes for training), customers didn’t want that component—they wanted something else.
- •Over 100 conversations revealed low pull for verification-only
- •Users claimed they didn’t need mistake-finding (whether true or not)
- •Humbling lesson: the stated pain wasn’t error detection in existing designs
- •The real pain emerged: producing the design itself
- 4:01 – 5:17
Finding the right problem via YC network: the custom Jetson Orin board insight
A serendipitous conversation with a YC friend surfaced a repeated market need: robotics companies wanted custom Nvidia Jetson Orin development boards. The team realized they could deliver this quickly, and that kernel became a direct path to a high-signal first deal.
- •YC environment increased surface area for high-quality problem discovery
- •Robotics customers repeatedly requested custom Jetson Orin dev boards
- •The problem was concrete, urgent, and matched the founders’ capabilities
- •This insight immediately redirected their focus toward board generation
- 5:17 – 6:20
First deal and validation: customers want the full solution, not a tool feature
Davide contrasts the earlier lukewarm feedback with the “instant yes” on the new offer. The key insight: companies don’t just want software that makes PCB designers a bit faster—they need the boards designed and delivered because talent and bandwidth are scarce.
- •Immediate positive response validated the pivot direction
- •Market constraint: not enough PCB design talent (especially early-stage)
- •Even large companies outsource due to bandwidth limitations
- •Diode’s value: generate boards end-to-end as the core deliverable
- 6:20 – 7:53
Turning services into a scalable product: “services as software” with verification ownership
Diana presses on venture scalability when the product looks like a service. Davide explains how AI-driven productivity plus domain expertise enables an 80/20 approach: Diode can deliver complete outcomes while owning a verification pipeline to ensure correctness and acceptance criteria.
- •“80/20” reality: perfection is hard, but practical solutions can scale with experts in the loop
- •Diode uses AI internally but verifies outputs before delivery
- •They take end-to-end ownership with customer acceptance criteria
- •Resembles traditional development services—just faster and more efficient
- 7:53 – 10:16
Reframing PCB design as a software problem: unlock LLM capability via code representations
Lenny details the core conceptual shift: represent schematics like code so models can express their latent electrical knowledge. Traditional EDA tools are visual and haven’t evolved much; by giving models a code-like medium, Diode harnesses software-era verification and automation techniques for PCB design.
- •LLMs already “know” electrical engineering rules, but need the right interface
- •Traditional PCB tooling is visual/graphical and slow to iterate
- •Convert board design into code to leverage software tooling and AI automation
- •Export back to visuals so humans can review without changing their workflow
- 10:16 – 11:52
Operational leverage: building internal tooling to ship far more boards
Diana highlights the speed: the founders produced over 100 boards in a short span. Davide explains how Diode’s internal division of labor (EE + software) and a custom software layer on top of tools like KiCad enabled multi-client throughput that would be impossible otherwise.
- •Two-person early team achieved unusually high PCB output
- •Davide leads EE work; Lenny builds software tools that multiply productivity
- •Use and appreciation of KiCad, plus a high-leverage proprietary layer on top
- •Internal tooling is treated as a product with the EE team as the first customer
- 11:52 – 12:22
Key technical bets: a schematic language designed for both humans and LLMs
Lenny explains the need for a new representation that LLMs can generate reliably while still being readable for human review. The language design is constrained by model familiarity (to reduce syntax errors) and by the practical needs of electrical engineers working day-to-day.
- •Novel challenge: language must work for both LLMs and human reviewers
- •LLMs perform better when syntax/structure resembles common programming languages
- •Goal: minimize model syntax errors and maximize natural code generation
- •Humans still need to inspect, edit, and trust the output
- 12:22 – 13:20
Infrastructure and security: air-gapped deployment, Rust compiler core, and browser-based review
To sell into conservative, regulated industries, Diode built an architecture that can run in high-security environments. Lenny describes an isolated Rust compiler as the core, with WASM bindings enabling local, real-time compilation and visualization in the browser, plus configurable modules that recompile instantly.
- •Security-driven requirement: support air-gapped environments (e.g., aerospace)
- •Core logic implemented as a tight, isolated Rust compiler
- •WASM bindings enable local in-browser compilation and visualization
- •Real-time recompile for configurable blocks (e.g., power modules) to speed review
- 13:20 – 15:28
What’s next: applying AI acceleration to the physical world—and who they’re hiring
Davide frames hardware as the next frontier for AI-driven productivity gains, with real-world reliability as the bar. Lenny describes the team profile they want: builders who enjoy open-ended research-like problems but can ship practical customer experiences for reviewing and handing off designs.
- •Vision: bring “10x” software-speed improvement to designing physical systems
- •Recruiting hook: hard, meaningful frontier problems in hardware + AI
- •Hiring: curious, experimental engineers comfortable with ambiguity
- •Also hiring for customer-facing interfaces for inspection, review, and feedback loops
