While LLMs will move the industry to well defined specs in English that are used to generate code and tests ranging from unit, integration and end-to-end, I don’t see the end of the role of humans in this process. I see the continued rise of the product engineer who is fully capable of interacting with stakeholders and customers while also having the time and energy to produce working software.

The singularity seems pretty far off and the current LLM techniques will change drastically over the coming years… Algol families exist but we’ve well beyond those abstractions.

One thing I’ve noticed is that LLMs will eventually run in to Syntax errors or other logical errors and halt on their own. I’ve been calling it the non-halting problem and intuitively it seems as though this will be a rather large hurdle to overcome!

There’s a certain amount of computational thinking required by a developer required to understand how the computer is interpreting code that seems missing from the capabilities of both LLMs and symbolic interpreters like traditional computers.