I wrote a piece for the Bollwerk blog about building an agentic workflow for company risk assessment. It walks through three iterations — from a rigid linear pipeline to an over-autonomous agent and finally to a focused feedback loop with parallel initial gathering — and the lessons that came out of each. Topics include PocketFlow, Modal, Langfuse for observability, and why tool proliferation hurts more than it helps.

Read it here: Building An Agentic Company Risk Assessment Workflow: Lessons Learned.