Skills To Learn

1. *Specifying problems, algorithms, architecture etc mathematically, formally and concisely**

As LLM-assisted coding grows, the exact implementation will become less important. The most crucial skill will be to abstractly understand the problem, or the algorithm at hand; and being able to represent it in such a way, maybe using pseudocode or math, and then letting a language model figure out an exact implementation.

This might be more true for research and science.

I think though there is a slight open question here: what will the models be better at implementing something for which we have a formal specification (say invariants, pre-/post-conditions, etc). Or if we have a pseudocode implementation? Maybe this delves into the nature of specification? what kind of specifications are best for LLMs to generate robust implementations?