[api]

It’s basically saying that good engineers are better at using the tool.

Because of course they are. This is true of every tool up to AI and it’s true for AI.

If you know what you’re doing you’re going to do a better job of getting the AI to produce good output. Isn’t that obvious?

[dijksterhuis]

> [performance variance between engineers] is now directly measurable [by measuring token usage]

… ie token use ~= productivity output

unfortunately measuring the number of tokens used is about as useful as a measure of productivity/performance as is measuring the number of times i hopped around on one foot last week (less time hopping ~= more time coding).

it’s just a measure of the number of tokens that an engineer used. it doesn’t necessarily mean that engineer is more productive. they might be doing more tokens because they ended up re-doing one minor feature a hundred times because they don’t understand the language / requirements etc. it could even be a negative relationship to productivity / performance!

pretty sure that’s what gp was getting at. see LoSC.

[delusional]

> It’s basically saying that good engineers are better at using the tool.

That's extremely reductive. It's perfectly possible for an engineer to be both a fantastic engineer, and bad at using AI. The opposite also exists, the great AI user, who is a terrible engineer.

The idea that these two "skills" are somehow 100% correlated or there's a causation link between them is completely unfounded. The person who becomes a fine engineer with AI, might have been an absolutely terrible engineer without it.

[clapthewind]

> It's perfectly possible for an engineer to be both a fantastic engineer, and bad at using AI

In this era, AI native skills are the major factor, alongside system design, which define who's fantastic. As an example, deep programming language or framework knowledge is a commodity now and not meaningful as a skill difference.

[delusional]

That entirely the wrong point to take away from my comment.