Anthropic says Claude Fable 5.2 designed three electronic devices, from circuit layouts and firmware to enclosures and parts lists, and helped engineers build prototypes that worked without circuitry fixes. If that process holds up beyond prototypes, AI-assisted hardware design could move from drafting help towards a practical engineering tool.
Anthropic Watch analysis
What happened
Anthropic and circuit-board company Diode Computers asked Claude to design a speakerphone, a voice-assistant hub and a handheld virtual pet. Anthropic says Claude produced specifications, schematics, board layouts and firmware, checked manufacturing constraints and parts availability, and revised designs when problems emerged. Diode manufactured the boards and physically tested the prototypes.
The examples include Claude rerouting crowded board connections, writing a grid-search router for long tracks, and checking that a board would fit inside its enclosure. Anthropic reports that Fable 5.2 scored 69% on its internal PCB Gym benchmark, compared with 25% for Fable 5, and 83.5% on the public BenchCAD 2.1 benchmark. Those are benchmark results reported by Anthropic, not independent evaluations.
Anthropic’s account of the electronics-design work includes the methods, examples and limitations.
Why it matters
This is a more concrete test of AI’s engineering abilities than producing a plausible-looking diagram: the designs were turned into physical boards, with firmware and enclosures, and Anthropic says they worked without circuitry fixes. The process still involved human engineers, who built and tested the devices.
These remain early prototypes, not certified or mass-production-ready products. They have not undergone electrical validation, antenna tuning or thermal testing, and Anthropic has a commercial relationship with Diode. That makes the work promising evidence of capability, not proof that engineers can hand over product design and walk away.
Our read
The impressive part is the chain of decisions: component placement, routing, manufacturing checks and mechanical fit all have to agree before a device works. Claude appears to have handled more of that chain than a tool that merely suggests a circuit. The useful next test is whether the approach scales to denser, more complex designs and survives independent validation.
What to watch
- Whether Anthropic tests Claude on denser and more complex circuit boards.
- How the designs perform in electrical, thermal and antenna testing.
- Whether independent teams can reproduce the results on hardware beyond these three prototypes.
Discussion spark: If an AI can design a working prototype but people still build and validate it, which parts of hardware engineering would you trust it to take on next?
Sources and evidence
- Claude can design electronic prototypes from start to finish – Anthropic (9 November 2026, 18:58 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.