AI coding created a new class of creators. I'm one of them.

Harald Reisinger — Wed, 01 Apr 2026 16:54:35 +0000

Something remarkable is happening. People who aren't developers
are building real software.

Not toy demos. Not no-code drag-and-drop pages. Real, compiled,
native applications with backends, databases, and streaming
architectures. AI coding tools — Claude Code in particular —
have made this possible.

I'm one of these people. I come from the world of ideas, products,
and design. I've never written Rust. I've never dealt with code
signing, DMG packaging, or IPC bridges. But over the past months,
I built CodeMantis [(https://codemantis.dev)] — a native macOS desktop application for Claude Code — using nothing but Claude Code itself.

Tauri v2. Rust backend. React 19 frontend. TypeScript. Real-time
streaming. Multi-process management. Under 30MB, launches in 3
seconds, uses 40-120MB of memory. Signed, notarized, and shipping.

A non-developer shipped a native Mac app with a Rust backend. That
sentence would have been absurd two years ago.

The Problem I Solved

Claude Code is the most capable coding AI I've used. But it lives
in a terminal — a scrolling wall of text where your conversation,
file changes, and command output all share one lane. I kept losing
track of what Claude had changed. Every developer I talked to had
the same complaints. They loved the AI. They tolerated the interface.

For someone like me — not a developer, but someone who wants to
build — the terminal wasn't just inconvenient. It was a wall.

What CodeMantis Does

CodeMantis wraps Claude Code in a three-panel interface: file tree
on the left, conversation in the center, activity feed on the right.

The separation matters. When Claude reads 15 files and edits 3,
you see the reads in the activity feed and the conversation stays
clean. No more scrolling through file contents to find the AI's
next message.

SpecWriter is the feature I'm most proud of. It's an AI
conversation that draws out the details you'd forget to mention:
loading states, error handling, validation rules, edge cases. It
produces a spec document with verification checklists. You hand
the spec to Claude Code, and the implementation quality jumps
dramatically.

Preview Browser shows your running app in a native window.
Describe a change → Claude implements it → see it live.

11 Project Templates scaffold new projects with a CLAUDE.md
that teaches Claude your project's structure from the first message.

Multi-AI Assistants let you chat with GPT, Gemini, or Claude
alongside your coding session for second opinions.

Who This Is For

Let me be honest. CodeMantis is probably a boring tool for senior
developers. They already have their terminal customized, their
workflows dialed in, their muscle memory trained.

CodeMantis wasn't built for them.

It was built for the founder with a SaaS idea and no engineering
team. For the designer who can see every pixel but has never
shipped a binary. For the domain expert who understands their
industry better than any developer could. For anyone staring at
a blank folder, wondering if this is really possible.

AI coding doesn't remove the adventure of building something. It
removes the barriers to starting one.

Try It

Free and open source (MIT license). macOS only for now.

Website: [(https://codemantis.dev)]
GitHub: [(https://github.com/codemantis-dev/codemantis)]
The full story: [(https://codemantis.dev/about)]
About me and my motivation: [https://codemantis.dev/about/harald]

Requires Claude Code CLI with a Claude Pro or Max subscription.

I'd love to hear from anyone who's building with AI — what works,
what's missing, what would make the experience better.

Forem: Harald Reisinger

AI coding created a new class of creators. I'm one of them.

The Problem I Solved

What CodeMantis Does

Who This Is For

Try It