On a Tuesday morning in early September, a USB-C breakout board designed itself in 47 minutes. No engineer drew a single trace by hand. The work happened inside Copperhead, an open-source AI agent that converts natural-language instructions into manufacturing-ready circuit boards, complete with verified schematics and bills of materials.
When founder Animesh Chouhan posted the project to Hacker News on September 10, it collected 242 upvotes and 113 comments within hours. The reception signaled something hardware engineers have been waiting for: the kind of AI-assisted workflow that software developers have enjoyed for years, finally arriving for the painstaking work of PCB design.
The idea is straightforward, if ambitious. Engineers describe what they want in plain language. Copperhead reads design docs, proposes changes, edits real KiCad schematics and board files, then runs automated electrical and design-rule checks before committing anything. If a proposed edit fails verification, the system rolls back to the last good snapshot. No unvalidated changes ship.
"It is not a chatbot that describes a circuit to you," Chouhan wrote in a July blog post. "It is a loop that leaves artifacts on disk."
That distinction matters in an industry where mistakes can mean scrapped prototype runs and missed launch windows. Circuit board design has resisted automation partly because the stakes are high and the tolerances unforgiving. Software can fail gracefully; a shorted power rail cannot.
Copperhead tries to bridge that gap by working within KiCad, the open-source electronics design suite already familiar to many hardware teams. The tool makes surgical edits to KiCad's s-expression file format, keeping diffs reviewable and refusing to touch uncommitted code. Engineers retain full version control.
The system operates in two modes. A "do" command iterates on existing repositories, reading specs and making targeted adjustments. A "create" command builds projects from scratch: it ingests a markdown design brief and outputs a validated specification, system architecture, parts list with rationales, an ERC-clean schematic, a DRC-clean layout draft, manufacturing exports, and a bring-up plan. That 47-minute USB-C breakout board run demonstrated the full eight-stage pipeline.
Chouhan Industries, the Bengaluru-based startup behind Copperhead, reports crossing 1,400 installs with over 300 active engineers. The CLI tool itself is free and Apache-2.0 licensed. The company monetizes through a cloud service priced at $49 per user monthly, which bundles hosted runs, a web viewer with live schematic rendering, one-click exports, and 200 inference credits per seat. Open-hardware projects released under CERN-OHL-S or OSHWA licenses get cloud access free within fair-use limits.
Team accounts add $199 monthly for CI bot integration, shared constraint libraries, SSO, and pooled credits. Enterprise pricing scales to self-hosted or VPC deployments, with custom annual contracts. The FAQ estimates model-token costs between ten cents and a dollar per change request; full project generation runs roughly five to ten times higher.
Copperhead is model-agnostic, supporting Claude, Cursor Agent CLI, Codex CLI, or direct OpenAI and Anthropic API keys. It requires Node 20 or later and KiCad 8 or later. Each run produces synchronized artifacts: specification files, BOMs with part justifications, schematic and board renders in SVG, DXF and STEP mechanical exports, firmware code with generated pin definitions, and a development plan. Transcripts land in a local directory with secrets redacted.
The competitive landscape is fragmented. Flux.ai offers an AI assistant called Copilot but requires migrating projects into its proprietary browser-based environment. Diode provides AI board design as a service; Copperhead claims Diode pricing is in the range of $10,000 to $50,000 per board, though Diode's site lists "Contact us for pricing." Quilter automates layout with physics-driven AI after schematic upload. DeepPCB ships a KiCad plugin using reinforcement learning for routing. Siemens released an EDA AI Agent last year for enterprise toolchains including Xpedition and HyperLynx.
Copperhead's pitch centers on staying open-source, working natively with KiCad files, and enforcing verification gates at every step. "What makes the loop trustworthy isn't the model," Chouhan wrote in a post submitted to the Antler Crackathon competition in July. "It's what the model isn't allowed to do."
That philosophy extends to a new benchmarking effort. Chouhan mentioned during the Hacker News discussion that the company was building Copperbench.org, an open-source standard for evaluating AI agents on verifiable hardware design tasks. The benchmark site launched in early September 2026 at version 0.1.0 with fixtures and tasks across three difficulty tiers, drawing on boards from Antmicro, Open Lighting, and a mikoto nRF52840 module. The leaderboards remain empty for now.
A third-party review from Compendia Labs, published September 8, tested Copperhead's core architecture on a minimal Debian virtual machine without LLM backends. The reviewer approved the design: "an OpenSpec-gated edit loop where every mutation is verified by ERC/DRC before commit."
Chouhan, who previously worked as an associate software engineer at JPMorgan Chase, founded Chouhan Industries in 2025. LinkedIn lists the company at two to ten employees. The homepage displays partner logos for Microsoft for Startups, Google for Startups, Activate, and Hugging Face, though these appear to represent program memberships rather than equity investment. The careers page advertises on-site Bengaluru roles for founding AI engineers, research interns, and forward-deployed hardware engineers.
Version 0.10.0 shipped August 26 with deterministic schematic drafting and improvements to the KiCad drafting stage. Whether Copperhead can sustain momentum beyond early adopters will depend on how well those verification loops hold up at scale, and whether hardware teams trust an AI to touch their production files. For now, the 47-minute breakout board stands as proof of concept—and perhaps a preview of how circuit design might look when the tools finally catch up to the code.
