Libraries are under-used. LLMs make this problem worse.

Libraries are under-used. Why? Briefly:

1. Writing code is more fun than reading documentation.
2. Dunning-Kruger effect leads us to understimate the complexity of the problem solved by the library we're considering.
3. Perverse incentives: libraries compete with big internal engineering projects that look good in a promo packet.

LLMs make this problem worse. Why? Less briefly:

Vibe coding is more fun than reading documentation. Shit, vibe-coding can be more fun than ordinary coding. You write a small prompt, get a ton of output, and get to revel in how long writing all of that code manually would have taken. You get to feel like we're on the cusp of a new era.

The problem is that the LLM's output just isn't as good as a battle-tested library. As we all know, LLMs don't follow our prompts perfectly, but even if they did, our prompts comes from our niave view of our problems. Niave prompt in, niave code out.

By contrast, library creators know the problem better than us. They have better prompts and the experience to shape an LLM's output into solid library code. For many complex problems, we should use their library, not an LLM directly.

Perverse incentives push us the wrong way here too. An engineer who uses an LLM to generate thousands of lines of code is a innovative leader who will help the company win in the new age of AI. "Promote them!," the managers say. I"m not saying that engineers consciously choose to use LLMs over libraries because of the hype. Just saying the incentives are there.