Specular

I've been thinking a lot that this era is the best possible time to learn the more fundamental pillars of software development, to figure out how the basics really work at a deeper level, and to aim for the top ranks of IT professionals while most of the industry is busy drinking Altman's Kool-Aid.

Maybe that's just my imagination, or maybe it's a fear that comes from watching people in my bubble let their skills rust by handing off their thinking to language models.

Of course that doesn't apply to everyone, but a number of people I know have completely shifted architecture and coding onto Claude, Codex, or whatever tool happens to be hot at the moment.

I'm not sure whether LLMs will ever be able to make use of depth of expertise in quite the same way, but depth of expertise is unlikely to go out of fashion anytime soon.

Developers who have spent years studying the field can learn to use today's most popular LLM in a matter of weeks.

Developers who have spent years doing development with nothing but an LLM are not going to gain a very deep understanding of the current state of the field and how things actually work in a week.

It will be interesting to see what companies that plan to build all their code with LLMs will look like five years from now.