Being an expert software developer - which Jarred Sumner indisputably is, having created Bun - doesn't automatically make you an expert on predicting the improvements in software development performance that LLMs enable. All of us - experts and amateurs alike - are in the process of figuring that out, in real time, around the world, right now.

Underestimating how quickly a non-trivial project will come together is an almost unheard of phenomenon. It used to invariably be the other way around, to the point that there are laws about it, like Hofstadter's Law, which says that projects always take longer than anticipated, even when accounting for the law itself. Or Fred Brooks' work, which puts limits on how much the development of software projects can be sped up.

The sane takeaway here is that if what's being reported is true (keeping in mind it's coming from a newly minted Anthropic employee), it implies an astonishing, unheard of improvement in software development speed, at least for certain kinds of tasks, enabled by LLMs.

To somehow twist that into "experts may not be as skilled and knowledgeable as they appear" or "not skilled in the tools they’re using" makes me think of the Charles Babbage quote, "I am not able rightly to apprehend the kind of confusion of ideas that could provoke such [an opinion]."