agents are unreliable verifiers
it's clear that deterministic checks with results fed back to agent is what's most effective in both time and money
especially in any sufficiently automated, software factory -like process
tbh. I wish more people didn't lean on the agentic loop 100%; instead, they should lean on @temporalio esq workflow engines as stages. There's an art in deciding whether to utilise the LLM loop vs. a deterministic workflow. You can do both; the magic is in combining and applying