🧩 Complexity comes easy to AI

Why is AI good at programming and coding tasks?

All coding tasks are written in highly structured programming languages.

The work is written down as specifications, samples, tests, documentation, tickets, and emails. On top of that, they live in a source control management system, e.g. GitHub, meaning the trail can be played back and forth as many times as needed.

LLMs excel in such scenarios, where a task leaves a rich trail behind. They do not need the job to be simple. They need the steps and expected output to be clear in order to learn from them. In software, there is a rich set of test suites which helps even more, since a paragraph of text can be interpreted but cannot easily be verified to be true. Feed in code and enough text on the expected output and how to verify it, and it's a cakewalk for LLMs.

Software developers did not merely create software; they also created a trail and archives of this work. Open Source did it in the open and in public.