"I expect promptfondlers will be fired en masse and struggle to find work during the AI winter, as their inability to work without a money-burning chatbot turns them into a drag on a company’s bottom line."
-
"I expect promptfondlers will be fired en masse and struggle to find work during the AI winter, as their inability to work without a money-burning chatbot turns them into a drag on a company’s bottom line."
-
"I expect promptfondlers will be fired en masse and struggle to find work during the AI winter, as their inability to work without a money-burning chatbot turns them into a drag on a company’s bottom line."
I made the mistake of reading the comments on LinkedIn when the paper that showed LLM-assisted coding made you less productive (but estimate your productivity higher) was published.
One of the justifications from the promptfondlers was to attack the methodology because, although the participants all had experience using LLM coding tools, they didn’t have experience with the precise one that they used in the study.
This suggests that the problem is far worse than the linked post suggests. It isn’t just that these people are useless without their chatbot tools, it’s that their skills with the, are not transferable. We saw this with the GPT-5 release as well: a load of patterns people had developed for working with GPT-4o stopped working and there was uproar.
I think you might have written about this before, but it’s a problem with nondeterministic tools. You don’t learn things that work, you learn things that have a high probability of working, but they are not based on a theory of the underlying system. So any small change to how the system works can make your techniques useless. So time spent learning to prompt an LLM is time spent both having your real skills atrophy and, in exchange, learning a skill that will be obsolete soon. No AI winter required.
-
I made the mistake of reading the comments on LinkedIn when the paper that showed LLM-assisted coding made you less productive (but estimate your productivity higher) was published.
One of the justifications from the promptfondlers was to attack the methodology because, although the participants all had experience using LLM coding tools, they didn’t have experience with the precise one that they used in the study.
This suggests that the problem is far worse than the linked post suggests. It isn’t just that these people are useless without their chatbot tools, it’s that their skills with the, are not transferable. We saw this with the GPT-5 release as well: a load of patterns people had developed for working with GPT-4o stopped working and there was uproar.
I think you might have written about this before, but it’s a problem with nondeterministic tools. You don’t learn things that work, you learn things that have a high probability of working, but they are not based on a theory of the underlying system. So any small change to how the system works can make your techniques useless. So time spent learning to prompt an LLM is time spent both having your real skills atrophy and, in exchange, learning a skill that will be obsolete soon. No AI winter required.
@david_chisnall @davidgerard I'm stealing "promptfondlers". What a wonderful word. Thanks.
