What I've learned about writing AI apps so far (link)
Citing Laurie Voss:
Is what you’re doing taking a large amount of text and asking the LLM to convert it into a smaller amount of text? Then it’s probably going to be great at it. If you’re asking it to convert into a roughly equal amount of text it will be so-so. If you’re asking it to create more text than you gave it, forget about it. The rest of this post is really just examples of this rule in action.
It’s easy to argue that as LLMs get bigger context windows and agent tools get more sophisticated that the ability to replace a whole human will show up: after all, LLMs are good at turning text into less text. How much text do you need to replicate everything a human knows? I don’t know but it’s a lot more than any LLM can handle right now. In the meantime, your attempts to replace humans with LLMs are going to suck. Let your app augment and accelerate a human, I’ve seen that work lots of times and be very effective.