Anthropic's Claude Fable 5.2 designed three electronic devices from scratch, and all three worked on their first build. The AI handled circuit board layouts, component selection, 3D enclosure design, firmware coding, and full manufacturing checks. Anthropic published the results on November 9 alongside hardware partner Diode Computers.

Specifically, Anthropic asked Claude to design and build a USB speakerphone, a voice-assistant hub, and a handheld virtual pet.

For each device, the company gave Claude a simple prompt describing the product and its core hardware requirements. From there, Claude wrote full specifications, drew schematics, placed components on four-layer circuit boards, and routed all copper connections. It also designed 3D-printable enclosures and wrote the firmware for each device. Diode then manufactured all three boards directly from Claude's design files without a single revision round. That outcome surprised the Diode team because factories and engineers typically require multiple adjustment cycles before production.

How Claude Fable 5.2 Achieved First-Time-Right Production

The AI's workflow began with a natural-language prompt that listed the target product, power budget, interface needs and any special performance criteria. Claude translated that brief into a formal specification sheet, including bill-of-materials estimates, component part numbers and cost targets. It then generated a complete schematic using industry-standard symbols and annotated each net with voltage and current requirements.

Next, Claude performed automated placement on a four-layer PCB. It evaluated thermal distribution, signal-integrity constraints and mechanical clearances, arranging components to meet the design rules of Diode's manufacturing process. The routing stage applied high-speed design principles: differential pairs for USB and Ethernet were length-matched, impedance-controlled traces were assigned, and via stitching was added to reduce electromagnetic interference.

When the layout became congested, Claude iteratively removed and re-routed connections, a process that normally requires several engineer-led design reviews. The AI also calculated acoustic resonance for the voice-assistant hub's microphone port, discovering a peak at 3.4 kHz. To move that resonance out of the speech band, it reshaped the enclosure into a wedge, shifting the resonance to 8.9 kHz.

Benchmark Performance Shows Rapid Progress

On benchmarks, Fable 5.2 scored 69 % on Diode's internal PCB Gym placement test, compared to just 25 % for Fable 5. Additionally, it hit 83.5 % on BenchCAD 2.1, a public benchmark for reconstructing circuit boards from engineering drawings. That score nearly doubles what Claude Fable 5.1 achieved on the same test.

BenchCAD measures an AI's ability to infer component placement, trace routing and layer stack-up from a set of engineering drawings. A score above 80 % indicates that the AI can produce a layout that a human designer would consider production-ready without major revisions. The PCB Gym test evaluates placement efficiency, component density and adherence to design-rule constraints, with higher percentages reflecting fewer manual adjustments needed.

What Sets This Generation Apart From Earlier AI Hardware Experiments

Earlier attempts at AI-assisted hardware design often stopped at schematic generation or required extensive human oversight for layout and verification. Claude Fable 5.2 demonstrates deeper reasoning across the entire product development chain. It prioritized high-speed routing before populating passive components, a strategy that preserves signal integrity in complex designs.

The AI also performed on-the-fly acoustic analysis, a capability rarely seen in prior models. By modelling the interaction between a microphone port and its enclosure, Claude identified a resonance issue that would have degraded voice-assistant performance. The automated redesign of the enclosure shows that the model can integrate mechanical considerations with electrical constraints.

Limitations and the Role of Human Engineers

Despite the impressive results, all three devices remain early prototypes that have not undergone electrical validation, antenna tuning, or commercial certification. Human engineers at Diode still handled all physical construction because Claude lacked direct access to assembly equipment.

Electrical validation includes functional testing under load, thermal cycling and compliance with safety standards such as IEC 60950-1. Antenna tuning is required for wireless modules to meet FCC or ETSI regulations. Commercial certification involves third-party labs and documentation that were not part of the AI-only workflow.

Therefore, while Claude can produce a complete design package, the final step of taking a prototype to a market-ready product still depends on traditional engineering expertise and regulatory processes.

Implications for the Electronics Design Industry

If AI models like Claude can reliably generate production-ready designs, the time and cost of bringing new hardware to market could shrink dramatically. Companies could iterate on form factor, component choice and firmware features in a matter of hours rather than weeks. This shift may lower barriers for startups and small firms that lack large in-house design teams.

However, the need for validation and certification means that AI will complement rather than replace human engineers in the near term. The technology is most valuable as a rapid-prototyping assistant, providing a solid baseline that engineers can refine and certify.

What to Watch for Next

Anthropic plans to release further updates to the Claude series, focusing on expanding the model's ability to interact with physical assembly tools and to perform automated testing simulations. Diode Computers has indicated interest in scaling the workflow to larger, more complex products such as IoT gateways and wearable devices.

Industry observers will monitor whether future versions can close the validation gap, either by integrating simulation engines or by providing design-for-test (DfT) guidelines that streamline certification. The next public benchmark releases from PCB Gym and BenchCAD will also reveal whether the performance gains are sustained across a broader set of designs.