A Love Letter to Claude Code
Claude Inbox Zero is a terminal-based email client that handles your emails with the help of Claude agents. It’s both a practical tool and an exploration of how we’ll work with AI agents in the future.
I built this for Anthropic’s “Built with Claude Sonnet 4.5” challenge, and it turned into something more interesting than I expected.
The UX Innovation
The compelling thing about this application is how it inverts the traditional workflow. Claude agents write email responses for you in the background, then dynamically show them for your review. If you want a revision, an agent goes off to handle that while you continue working on the next email in your queue.
There’s something subtle happening here: the agent is responsible for replying to emails, and it’s prompting you for feedback. It shows you context that might be helpful. The tables are turned.
I think we’ll work with agents a lot more like this in the future.
Building It
The development process was a collaboration with Claude Code itself:
- Wireframing: I vibe-wireframed the interface with Ink to find the right controls and feel
- SDK Implementation: Asked Claude Code to implement the Claude Agent SDK, using the some SDK demos as reference. It gave guidance on multi-turn conversations, resume support, and other patterns
- Learning from Others: Used the open source Gemini CLI to identify UX patterns we were missing, like the login flow
- Rapid Iteration: Vibe coded the whole thing quickly—it’s still a bit sloppy but functional
Try It Out
The app is open source on GitHub and available to play with. Fair warning: an email client is probably the easiest way to test prompt injections, so use with caution.
This is what happens when you give developers powerful tools and let them build what they want. More to come as I clean this up.