2025.stateofhtml.com

we're all "random people" to other random people…

This was a huge amount of work for everybody involved, with @lea.verou.me staying up so late to work on this that she more or less ended up matching my own work hours despite living on the opposite side of the globe…

So I hope you enjoy the results!

2025.stateofhtml.com/en-US/

Finally, we realized we had so much data that we decided to make things a little bit easier to digest by calling out one specific feature per section as the one feature you should learn (or at least remember) this year.

The best part is that you *also* get access to respondents' original, raw answer for any topic–and can even cross-filter those down further by an additional secondary topic if you want.

This in turn pushed us to rethink how charts could be made to show nested data, ending up with this file tree/bar chart hybrid.

Lea and her colleague Dmitry Sharabin also developed a LLM-based workflow which let us extract a LOT more datapoints from freeform questions, going from 70-80% of answers analyzed previously to 95%+.

LLMs may not always be the answer, but they're pretty good at detecting patterns in messy data.

Somehow I feel like I know what prompted this hot take…

Pretty soon people will be laughing at the old-fashioned people who insist of using a smartphone instead of just getting an OpenAI Brain Implant™