Summary
Indie app developer Chris Raroque shares what he learned after announcing Moonjar, a screenshot tool built for AI agents that drew 1,800 waitlist signups. The tool lets an agent capture screens for something like a help center, lay them out on a canvas, and apply edits such as highlighting areas with arrows.
When the first three testers struggled, with two failing to get past setup, Raroque ran 30 onboarding calls in two weeks, watching users think out loud while he stayed silent. He describes the friction points and “magic moments” he found, which analytics alone couldn’t reveal.
Key changes included replacing a daunting wall of instructions for installing skills and MCP with agent-specific setup, taking advantage of one-click MCP and skill installs supported by Claude Code, Cursor and Codex, and even adding an artificial delay so users believed the install had actually worked. He also moved permission requests to the moment they’re needed.
Raroque explains how he shipped around 933 updates by dictating fixes into Claude Code during live calls using Whisper Flow, which also took meeting notes. The video closes with an honest look at an unresolved question about where the product stands. It’s a practical read for anyone building software for AI agents.
📺 Source: Chris Raroque · Published November 10, 2026
🏷️ Format: Workflow Case Study
