
James Bennett‘s blog describes him as “a software developer living in California”.
Which is all I know about him.
But on that blog today he has a pretty good piece up called “Let’s talk about LLMs“.
His basic interest in the piece is the claimed “productivity” gains from LLM-assisted software development and whether they are (or even can be) as significant as their boosters claim.
Spoilers: he decides that no, they are not and cannot be.
So, no surprises there.
But he makes a really strong and thoughtful case for it, which is very much worth reading.
And he gives the strongest case for LLMs, too, basically ignoring their fundamental inaccuracy, their origins in mass plagiarism, their drag on the processing power and bandwidth of the entire Internet and their massive thirst for equipment and electricity.
No, he takes the rosiest view of LLM-assisted development’s utility and argues that even given all that there’s no way they could give even a single order of magnitude improvement to developer productivity.
Harsh, but fair.
He’s much more kind about it in the essay than I’m being here, though.
You should read it.
- You Should Read This: Let’s Talk About LLMs - 2026-04-09
- Make It Again: The 25th Amendment Is Just Bad - 2026-04-08
- American Hubris: Not As Rich As Croesus, But Much Dumber - 2026-04-07
