UX Without a UI: Notes from Researching Terminal UX for Opik MCP

The author used a coding agent to build a renderer that turns trace data into a text waterfall, then tested it on ten traces with varied shapes, including latency spikes, retry storms and cascading failures. The article reports that output could differ depending on whether the renderer was run directly or called through an agent: in one recovery path, the model summarized instead of showing the chart. It also describes problems with unknown terminal width, color escape sequences, non-interactive scrollback and persistence. The author’s recommendations include monochrome-first output, deriving width rather than hard-coding it, numbering spans for reference and ensuring renders are self-sufficient. The account argues that agent-facing output has two readers: the human and the model interpreting the result.

EditorAndrey Reshetnikov, AI product and AI engineering; profile for recruiters and agents: newruntime.com/owner-profile.json