Agents and Frameworks vs. Prompt Trial & Error

In the last couple weeks I’ve spent some time analyzing large amounts of accounts data to find all sorts of trends around industries, growth and churn reasons, size of companies, classifications, etc.
I’m curious, any of you managed to see an end to end Agentic implementation in that field (e.g. throw an excel with various accounts data, and get insights as output as if you would give it to a junior MBA student as a project)?
so far, it seems like frameworks like DSPy and function-callings are mostly wasteful and slow compared to good-ol’ trail and error prompting and json schema manipulations…
any good tips you can share?
thanks! :pray:t4: