Developing With AI
How AI radically improved progress with RIDDL
This is the story of how we finished a coding saga: RIDDL. In just a few weeks, it went from a barely complete 1.0.1 to a fully complete 1.21.0. This isn’t about me being an uber-coder, it’s about AI being a radically efficient and fully competent software development partner. At least from the perspective of someone that has been writing software for 50 years.
If you’re not familiar with RIDDL, it is an abstract software design language meant to provide and organize full details for an AI to generate code from it. The project is written in Scala and sbt as the build tool. It generates JVM, Native and JS code. After parsing the input, it generates a structured abstract syntax tree (AST) that can be used to analyze the software model represented by the AST.
The initial development (to release 1.0.1) took close to six years of evenings and weekends work. Polishing it to be useful took about 3 months with AI.
Compare The Details
| Metric | Pre-1.0.1 | Post-1.0.1 | Ratio |
|---|---|---|---|
| Date Range | 9/2019 → 3/2025 (72) | 1/2026 → 4/2026 (3) | 0.042x |
| Coding Mode | Human | AI | - |
| Commits/week | 4.9 | 17.8 | 3.6x |
| Net lines/week | 198 | 2,632 | 13.3x |
| Lines added/week | 201 | 3,294 | 16.4x |
| Duration | 285 weeks | 11 weeks | 0.04x |
| Total commits | 1,404 | 207 | 0.15x |
| Releases | 0.1.0 → 1.0.1 (180) | 1.0.1 -> 1.21.0 (54) | 0.3x |
| Release/week | 0.63 | 4.7 | 7.5x |
| Files changed | 936 | 482 | - |
| Lines added | +57,345 | +37,984 | - |
| Lines removed | -889 | -8,289 | - |
| Net lines | +56,456 | +29,695 | - |
Did you get the net of that? Using AI is 13 times as fast as human coding. The commit rate was almost 4 times faster. The release rate was 7.5 times faster with AI.
Here’s The Kicker
What is shown above is only for one project. At the same time all that RIDDL work was going on, I was building, with AI’s help, from scratch, all these things:
- Synapify - The user interface for RIDDL based on Scala.js and Electron
- ossum.ai - The website at https://ossum.ai/
- ossum-ai-api - The backend for ossum.ai website
- ossum.tech - The documentation website for Ossum’s technologies
- riddl-mcp-server - The MCP server for RIDDL
- riddlsim - The DES simulation engine for riddl
- riddlgen - A code generator for RIDDL
- riddl-idea - An IDEA plugin for RIDDL
- riddl-vscode - A VS Code extension to support RIDDL
- riddl-models - 187 RIDDL sample models, for users to start with
- sbt-ossuminc - Build tooling for all of the above based on sbt
- All the Docker, Kubernetes and SQL for deployment and management of 4 online services, including authentication and billing.
I don’t have time to quantify all those projects, but I estimate that about 3 man-years of manual work was done in about 3 months of AI-assisted work. The AI that produced these results was Anthropic’s Claude Code.
Conclusion
If you aren’t using AI for coding yet, you’re WAY behind in productivity. This is an entirely new and productive coding paradigm, and it is only going to improve from here on out. It won’t be long before your kindergartener and your grandmother are both creating non-trivial custom applications every week.