Claude Code adds a /design command for terminal mockups, and NVIDIA puts code agents to a prompt test
Anthropic added a /design command to Claude Code and renamed its Workbench to Playground, while NVIDIA showed prompt detail changed the code but not the physics.
Artificial Intelligence··Midday
The /design command drafts interfaces in the terminal
Anthropic has introduced a /design command in early preview for Claude Code, allowing developers to generate user interface mockups directly in the terminal. The command produces shareable Artifacts that match the project's existing codebase style. These mockups are presented as artboards that developers can select, edit, and refine before writing actual code. The feature requires manual saving to move designs into the build stage, bridging the gap between prompting and visual interface iteration.[1]
Workbench rebranding
Alongside the terminal updates, Anthropic's release notes confirm that the Workbench in the Claude Console has been renamed to Playground. The rebranded interface supports all Messages API parameters and provides the full SDK request next to the API response, removing the need for developers to guess the request structure. The platform also added new templates that demonstrate specific features, including code execution and web search capabilities.[2]
Prompt precision changed the code, not the physics
In materials science, NVIDIA tested AI coding agents across three workflows using H200 GPUs to see how prompt specificity affects the output. Running the ALCHEMI Toolkit, the agents built 45 simulation pipelines. While the level of detail in the prompt changed the structure and computational cost of the generated code, all 45 scripts successfully computed the same physical quantities. The tests showed that liquid-lithium self-diffusion and oxygen adsorption converged on the correct physical results regardless of the specific code path.[3]
