Six interfaces for understanding coding agents

Sun, 04 Oct 2026 00:00:00 +0000

A coding agent can read files, run commands, and delegate tasks faster than a person can follow its conversation. Watching the tool calls arrive is useful, but supervising the work also means understanding who is responsible for each task, which activities take time, and how to guide the next run.

For Assignment 1 in CS2680, students built web interfaces around Claude Code’s headless mode. The assignment called for streamed tool calls and their status, run costs, follow-up messages in the same session, nested subagent activity, and an outline of the run.

Across the 68 demos we reviewed, a common design emerged: a conversation with collapsible tool-call cards, an outline alongside it, and a cost display. The six projects below explore additional ways to organize that information. Each makes a different part of the agent’s work easier to understand.

You can explore all six in the Mads Lens collection. Each app includes recorded runs or a demo mode, so you can try its interface without installing Claude Code or consuming Claude usage.

Subagents on CS2680 Student Blog

Six interfaces for understanding coding agents