As part of her design workflow, Meaghan Choi sometimes provides engineers with a draft pull request created with Claude Code instead of a traditional Figma mock-up.

Meaghan Choi uses Claude Code to understand a codebase's implementation, system prompts, and architecture as a pre-brainstorming step for new features.

Meaghan Choi, a Design Lead at Anthropic, spends approximately an equal amount of time working in Claude Code as she does in Figma.

Meaghan Choi believes that AI coding tools are making roles such as designer, product manager, and engineer more fluid by giving non-engineers the ability to code.

Meaghan Choi has configured her personal `claude.md` file to instruct Claude Code to reference the existing component library and adhere to design system best practices before finalizing implementatio...

To provide visual feedback for UI changes, Meaghan Choi's workflow involves exporting a design mock as a PNG and dragging it into Claude Code to leverage its vision capabilities.

Claude Code supports multiple, hierarchical `claude.md` files, allowing for layered context and instructions.

Anthropic is actively working on training Claude to be proficient at research-oriented coding, in addition to its capabilities with production code.

Employees at Anthropic are frequent users of Claude Code, serving as the primary 'dogfooders' for the product.

Anthropic recently released a feature for Claude Code called 'agents' (or 'sub-agents') which are programmable agents that can run in parallel.

Claude Code can connect to MCP (Multi-Component Protocol) servers, with common examples including a Figma MCP used by designers and a Playwright MCP used for application previews.

At Anthropic, many designers were already coding and submitting pull requests even before the direct integration of Claude Code.

Claude AI's artifacts feature can be used for 'zero to one' product exploration without needing the full Claude Code tool.

Claude Code has a 'background bashes' feature, activated by Control+B, that allows ongoing processes like a dev server to run in the background while the user continues to interact with Claude.

Claude Code supports a personal `claude.md` file that provides global instructions for a user's instance across all repositories, separate from project-specific `claude.md` files.

The 'learn mode' feature in Claude Code originated as a custom slash command built by an individual team member at Anthropic, which was later developed into a full product feature.

Anthropic's product development philosophy for Claude Code involves releasing features to the public early to gather community feedback for continuous updates and improvements.