Understanding LLM agent architecture for groundwork-ai

Anthropic's LLM Agent design patterns#

I got a lot of value in finding clarification on what LLM patterns to use to build each section in groundwork.ai, I have been digging in Routing workflow to generate 4 different cases (market, use case, process model and timeline) from a single generic SOT output.

Turns out this can deeply affect how solidified my responses can get. I am thinking of not only using LLM models for the pricing criteria, but I'm also looking at the LLM design patterns placed under the hood that I'm gonna use to generate prompt responses. I think I'll stick to one cheap model and try to experiment having different workflows under the hood to make the pricing.

The higher the price, better the pattern, that's what I am thinking for now.

Reference: https://www.anthropic.com/engineering/building-effective-agents