When we build products based around agents and language model based decision making, the differentiator in the product itself, the thing that separates your solution from someone just using ChatGPT directly, is your ability to produce reason, produce value and be helpful using heuristics and be deterministic. This shows the user that whoever built the solution already has built in understanding of your problem (you didn’t specify all the rules, they were already there), and it increases trust that the service will give you the same result as another person. Apply language models intentionally, only where it genuinely helps, and only when it can solve a very clearly defined problem for you, otherwise don’t.

Content ~ ( Solution ) - Interface

(Use LLMs to fix the dirty content - don’t use it to write prose about the facts)