Anthropic reported on November 9, 2026, that Claude Fable 5.2 had designed three gadgets that Diode engineers then built: a USB speakerphone, a home voice-assistant hub and a handheld virtual pet. For hardware startups and makers, the experiment explores AI’s ability to coordinate electronics, casing and software.
Anthropic says all three devices worked without electrical modifications. These are results reported by the participants, not independent replication. Anthropic discloses a commercial relationship with Diode.
What Claude designed—and people built
According to Anthropic’s launch account, Claude selected available components from US distributors, laid out four-layer circuit boards, designed enclosures and wrote firmware—the software that runs the devices. Each board design was ready for manufacture in under a day, the company says. That timing does not cover the complete build and validation process.
People supplied requirements and occasional short prompts. Claude had no connection to fabrication machinery or test equipment: human engineers built and tested the devices.
The study describes a practical assembly problem: the speakerphone’s mute switch obstructed insertion of its board. Claude modeled tilting the board by about 25 degrees to fit it inside the casing, then added that step to the instructions.
The significance is the coordination between design tasks: a circuit drawing alone would not resolve how to install the board. And needing no electrical fixes after fabrication does not mean the design reached that point without revisions.
Working prototypes are not finished products
Anthropic says the devices have not undergone electrical validation, antenna tuning, thermal testing or certification and were not designed for mass production. Long-term durability remains unknown. Three working examples cannot establish a general success rate or commercial reliability.
Fable 5.2 is available through the Claude API, but model access alone does not establish access to the study’s complete design setup. The available evidence also does not establish total project costs or engineer-hours saved.
For a hardware team, this supports considering a bounded prototyping trial, not dispensing with engineering review. RohitAI’s guide to evaluating AI-assisted engineering explains how to judge accepted designs and total effort before extending a trial.