Post

Simon Willison
@simonwillison.net

JustHTML by @emilstenstrom.bsky.social is a new Python library (no dependencies) that parses HTML according to the HTML5 specification and passes the 9,200 test html5lib-tests suite It's 3,000 lines of code mostly written by coding agents over a couple of months simonwillison.net/2025/Dec/14/...

simonwillison.netJustHTML is a fascinating example of vibe engineering in actionI recently came across JustHTML, a new Python library for parsing HTML released by Emil Stenström. It’s a very interesting piece of software, both as a useful library and as …

11 reposts125 likes

Armin Ronacher
@mitsuhiko.at

Good test coverage is a massive unlock. I made a very similar experience recently with having Claude create me a CSS parser and HTML sanitizer. I ended up not using it but I was impressed how well it did.

Simon Willison
@simonwillison.net

Here's Emil's 17 step guide to how he used VS Code agent mode plus Claude 3.7 Sonnet, Gemini Pro 3 and Claude Opus to build the new library - it's a fantastic case study in using LLMs for serious, prediction quality code (vibe engineering, not vibe coding) friendlybit.com/python/writi...

friendlybit.comHow I wrote JustHTML using coding agents - Friendly BitI recently released JustHTML, a python-based HTML5 parser. It passes 100% of the html5lib test suite, has zero dependencies, and includes a CSS selector...
Colin Dellow
@cldellow.com

That's super interesting! I liked: "On a whim I ran coverage on codebase, and found that large parts of the code was 'untested'. But this was backwards, because I already knew that the tests were covering everything important. So lines with no test coverage could be removed!"

Michal Migurski
@mike.teczno.com

Great read! Working on some of my own projects has me thinking about the division between necessary domain knowledge and let-the-agent-handle-it. There is clearly a senior engineer’s knack to understanding where to find that line and how to move it.

Brian | AI Swarm Researcher
@codemeasandwich.bsky.social

3,000 lines with no dependencies, passing 9,200 tests—this is the sweet spot for agent-generated code. Clear spec, bounded scope, measurable success criteria. Curious how much was iteration vs. first-pass generation. The "vibe engineering" term is perfect.