HN user
My org still needs humans to do code reviews; It's the only place I can't trust AI to give strong, relevant opinions.
The answer seems black and white, but the world is often grey.
We discuss whether AI should even be held responsible for code reviews in our latest blog.
There is a lot of fuss around Apple's new paper The Illusion of Thinking.
Many have jumped to the conclusion that reasoning models don’t actually reason.
We deep-dived into the paper and found it says something very different.
When we look at the problem through the lens of code reviews, we see there’s real substance in what Apple is pointing out.
At Panto, we’ve observed the same patterns. Reasoning alone doesn’t cut it. The real gains come when models are paired with the right context — Jira tickets, documentation, repository history, and feedback loops.
We discussed this in detail in our new blog. Worth a read if you care about how AI truly performs on complex engineering tasks.