AI notice
The Scenic Route, both the app and this website, was built with the help of AI. This page explains how, and the rules I've kept to.
How it was made
I'm one person building this in my spare time. Most of the code, and first drafts of much of the writing here, came from Claude Code, an AI agent made by Anthropic that can write, run and test software on my computer.
The way we work is human-led and agent-assisted. I decide what to build, often from something I noticed on a real drive. Claude Code researches it, writes the code and tests it on an Android emulator and on copies of the server. Designs are mocked up first and only built once I've approved them. I then test each build on my own phone, on real roads, and send back what's wrong. Every build is written up: what changed, and what was and wasn't tested.
What it made possible
Some of this I couldn't have done alone in the time. One example: the app plans routes offline with GraphHopper, an open-source routing engine built for servers. Its routing rules are normally turned into code while it runs, which Android can't do. Claude Code rewrote those rules for the phone, then checked them against the originals on hundreds of thousands of road sections. They matched every time.
By the numbers
All of this was built in about six months of my spare time.
- About 80,000 lines of code: the Android app (around 52,000 lines of Kotlin), its server, the tools that build the map and routing data, and this website.
- 383 builds so far. Problems found on a drive were often fixed within days, sometimes the same evening.
- More than 40 screens, covering everything on this site, from offline routing to the Convoy radio.
- About 450 hours of my own time, testing on real roads included.
Done the traditional way, I estimate that's 6,000 to 10,000 hours of work, or three to five years for one full-time developer. At typical UK agency rates of £500 to £800 a day, it would cost somewhere between £400,000 and £800,000.
So each hour I put in did the work of roughly 15 to 20. I decided what to build and checked that it worked; the agent did most of the writing and testing in between.
The AI side came to roughly 15 to 20 billion tokens. Paid for at the standard API prices, that's around £3,000 to £5,000. On a Claude subscription, I've paid about £108.
It's astonishing how far AI has come, and I look forward to working alongside AI agents for a long time yet.
Using it ethically
- A person decides. I approve every feature, design and release, and every change to the live server. The AI publishes nothing on its own.
- Checked, not trusted. AI-written code is tested, and anything that affects your driving is tried on real roads before it reaches you.
- Your data stays out of it. The app has no AI in it and sends nothing to an AI service. Passwords and keys are never shown to the agent.
- Credit where it's due. Map data, photos and routes from other people are used under their licences or with permission, and credited.
I'll keep using AI the same way as the app grows, with me deciding, checking, and answering for what ships.