Anthropic works on designing and prototyping three electronic devices with Claude Fable 5.2

Anthropic worked with Diode Computers to have Claude Fable 5.2 design three devices: a speakerphone, a hub for a voice assistant and a portable virtual pet. According to Anthropic, the model considered circuit-board layouts and manufacturing constraints, modeled the designs in CAD software, and then directed human engineers to build prototypes.

According to Anthropic, Claude worked not only on circuit-board design but also on designs that accounted for manufacturing constraints, and modeled them in CAD. Human engineers were directed to handle prototyping. When evaluating support for electronic-device design, key questions include whether proposed designs meet real manufacturing conditions and which steps still require human involvement.

The point to watch in this trial is Anthropic’s account that one model handled multiple stages of the design process, from selecting electronic components to board layout, checking manufacturing constraints, modeling in CAD and directing prototyping. The work involved three types of devices, making it broader than a single circuit-board task. However, this is a report on design challenges set by Anthropic and Diode Computers; it does not show that a model autonomously completed the entire electronic-device design and manufacturing process. In particular, human engineers built the prototypes. In earlier trials, the model reportedly struggled to lay out circuit boards reliably and to ensure compatibility between components. The difference in the latest account is that the evaluation expanded from using knowledge about individual components to design work that combines multiple elements. At the same time, the description provided does not establish how much Claude did for the three devices in areas Anthropic identified as necessary for electronic-device design, including modeling mechanical parts, writing firmware, identifying available components and understanding factory capabilities. It would be unwise to interpret the account as showing that Claude carried out all of these tasks. For a link to manufacturing, the boundary between software design and a physical product is whether a model’s proposed design fits a factory’s equipment and the availability of components. Anthropic said the model considered manufacturing constraints in this trial, but did not specify which constraints it was given or how it determined whether a design met them. To assess the potential for use in design processes, important verification points include not only the model’s output but also the amount of human revision and whether the prototypes worked after they were built. The fact that three devices were designed does not, by itself, establish the quality or reproducibility of the results. The description provided gives no comparative figures for whether prototypes were completed, whether the devices performed their intended functions, or the time, cost and number of revisions involved in design and production. The next question is which tasks Claude actually performed from design through prototyping, and where humans made decisions or revisions. Without an evaluation by process stage, the results cannot be linked to productivity in practice or the scope for use on manufacturing floors.