Launches, releases, funding and papers about agents doing hardware work.
Newest first, as we log them.
Filtered by
55 d ago
The atopile team published EEBench, a benchmark that grades agents on 13 circuit-design tasks written in atopile. The harness builds each design and runs SPICE checks at worst-case tolerance corners. The post says Claude Opus 5 led V1 at 61.6%. One failed design used a 22 uF capacitor that gave about 11.4 uF under bias. V1 is simulation only.
2 mo ago
Anthropic opened a research preview of the Model Hardware Standard, a shared driver spec for agents that operate lab and manufacturing devices. Each device describes its settings and enforced safety limits, and agents reach it over MCP, a CLI or code. In a QuEra case study, an agent-improved laser relock script worked in 99.3% of 700 trials.
2 mo ago
Arduino released App Lab 0.10 with Agentic Mode, which connects a bring-your-own-key model to an UNO Q project through MCP and the App CLI. Claude is the first supported provider. The agent can create and edit files, run and stop the App and read errors as they happen. The post does not show it reaching the board's pins directly.
2 mo ago
Dave Vandenbout ran his datasheet-to-symbol-lib Claude skill on 100 parts with 20 or more pins and diffed the results against KiCad's stock symbols. With every difference checked against the datasheet, generated symbols had a 2.77% pin error rate and the library 21.92%. The README notes that the same agent generated the symbols and judged both sides.
2 mo ago
Flux added a public MCP server to its PCB design tool in its summer update. External agents such as Claude, Cursor and Codex can query a board's live context and drive design changes. The post gives firmware headers generated from actual pin assignments and net names as one use. The update also adds a lower-cost chat mode.
3 mo ago
Researchers had coding agents write robot behavior trees from a contract served by a robot-side MCP server. The contract lists the robot's skills, allowed operators and optional templates, and a gate validates each tree before it runs. The paper reports tests of Sonnet 4.6 and Gemma4:31b on 110 simulated tasks and 14 on a Husarion Panther running Nav2.
4 mo ago
Zhihan Zhang and colleagues, mostly at the University of Washington, published Embedded Arena, a benchmark that scores agents' model and firmware edits on microcontrollers. Boards include a MAX78000, an ESP32-S3 and an STM32N6. The paper reports Claude Opus 4.7 and Gemini 3.1 Pro failing every deployment without hardware feedback; with it, the first success came within 3 iterations.
5 mo ago
Espressif published a recap on 14 May of ESP32 projects from Anthropic's Code with Claude event, where each attendee got an ESP32-S3 M5Stack Cardputer. Attendees described a program in plain language, and Claude Code wrote and flashed it. The post also covers Claude Desktop Buddy, a BLE reference design on the M5StickC Plus, and Espressif's own ESP Desktop Buddy library.
