timo way

Building Lunaway With a Room Full of Claudes

Lunaway is coming soon. It’s an astrology travel app: aligned travel for children of the sky. You tell it who you are under the stars, and it helps you figure out where to go and when.

This post isn’t the launch. It’s about how it’s getting built, because the way it’s getting built is the most interesting thing I’ve done with software in a long time.

One project, many sessions

I didn’t build Lunaway with one AI assistant in one long chat. I built it with a room full of them.

Each piece of work got its own Claude Code session with its own slice of the problem and its own branch:

  • Design passes. One session owned the look and feel. It went screen by screen, tightened spacing, picked type, and came back with a pass I could react to.
  • Security review and fixes. Another session did nothing but try to break things. It read the code like an attacker would, wrote up what it found, and then fixed it on its own branch.
  • Landing copy. A session that only cared about words. What the app is in two lines, what the buttons say, what you read before you decide to care.
  • An icon set. One session drew the icons. Consistent stroke, consistent grid, one family instead of a junk drawer of borrowed glyphs.
  • AI guide safety tuning. Lunaway has an AI guide inside it, and a guide that talks about your chart and your travel plans needs guardrails. One session’s whole job was tuning how it answers, what it won’t say, and how it handles the hard questions gracefully.

None of them had to hold the whole app in their head. Each one went deep on one thing.

The integration session

Then there’s the session that ties it together. One integration session’s job was to merge everyone else’s work: pull in the design pass, pull in the security fixes, pull in the copy and the icons, resolve the places where they stepped on each other, and keep the main branch building.

That’s the part that surprised me. Parallel work is easy to start and hard to land. The integration session is what turned five good branches into one app instead of five arguments.

What I actually did

My job changed. I stopped typing code and started doing the things a founder is supposed to do: deciding what Lunaway is, saying yes or no, and reading results.

A design pass came back and I’d say “warmer.” A security write-up came back and I’d say “fix all of it.” The copy came back and I’d cut half of it. Each session kept moving while I was away, and when I came back there was something real to look at.

It felt less like programming and more like running a small studio where everyone works fast and nobody sleeps.

What I learned

  • Small slices beat one giant brain. A session with one clear job does better work than one session juggling everything.
  • Someone has to own the merge. Without the integration session, parallel work just becomes parallel mess.
  • Safety is its own job. Tuning an AI guide is not a checkbox at the end. It got its own session from the start, and it shows.
  • Taste is still the bottleneck. The sessions produce. Deciding what’s good is still on me.

Lunaway is coming soon. If you’ve ever picked a trip because it felt right in your bones, it’s for you.