A Google engineer built syntaqlite — a full SQLite devtools suite — in 250 hours over 3 months using AI coding agents, documenting exactly where AI helped and hurt.
A senior engineer at Google who maintains PerfettoSQL (a SQLite-based query language) spent 8 years wanting to build professional-grade devtools for SQLite. Using AI coding agents over ~250 hours across evenings and weekends, he shipped syntaqlite — a formatter, linter, and editor extension suite. He published a detailed post-mortem covering both AI wins and failures, backed by a project journal, coding transcripts, and commit history. The project is open-sourced on GitHub.
This is one of the most honest technical post-mortems on AI-assisted development published by someone with real engineering credentials. The key finding isn't 'AI is amazing' — it's that AI agents compressed the unglamorous scaffolding (parsers, formatters, boilerplate config) enough to make a solo 250-hour project viable where it wasn't before. The critical failure mode documented: leaning on AI for the 'soul' of the software produces brittle code that collapses under user contact.
Pick one side project you've abandoned due to boilerplate overhead — specifically something with a parser, formatter, or CLI scaffolding problem — and run a 2-hour AI agent sprint on just that component using the syntaqlite repo as a reference implementation.
Suggested exercise · not independently tested. Access requirements and costs may vary.
Go to claude.ai and open a new conversation
Paste: 'I want to build a formatter/linter for [your DSL or config format]. Here is the grammar: [paste 10-20 lines of your syntax]. Generate a tree-sitter grammar stub, a basic AST walker, and a formatter that normalizes whitespace and keyword casing. Output each as a separate file.'
Review the output — the grammar stub and formatter boilerplate will likely be 70-80% usable; flag the parts that require domain knowledge the model got wrong
Tags