# What improved this year: the model, or us?

Published 15 July 2026

[Originally posted on LinkedIn](https://www.linkedin.com/feed/update/urn:li:activity:7483112297532665857/)

When Fable dropped, the feed lit up with "*wow, what a beast, look at the vuln it caught for me*," and I got worked up right along with everyone else. 🤯 Then it occurred to me that we started getting worked up way too late.

For some reason we all got excited at Fable, that it is such a huge deal, while the previous models were already doing the job just fine. We just didn't know that, because we weren't really watching them. 🤔

AI could already write code fine a year ago, it just did it 3 times out of 10, and now even Fable produces relevant code 7 out of 10. The new AI still can't produce code a human would call quality, and that hasn't shifted with Fable, or with the release before it.

The only thing that really moved was us: **learning to describe a task in a way the AI chews and interprets exactly as you intended, and learning to read back what it produced**. For any team building on this, that describing-and-reading skill is where the year actually went, and none of it ships in a release note. 🙃

By the way when another engineer hands you code, you think the exact same thing: "*what did you even write here*?". A person just writes slowly, so you have time to figure it out, while the AI writes fast, so you don't. That's the whole difference, and it has nothing to do with how smart the model got, only with whether you can keep up with reading it.

So what actually improved for you this past year: the model, or the way you hand it work? 👇