Florian Juengermann¶
Co-founder and CTO of listen-labs, an AI startup that turns hundreds of qualitative interviews, surveys, and focus groups into structured, traceable insights. Appeared on langchain's Max Agency podcast with harrison-chase in April 2026 (source).
Has rearchitected Listen's agent stack multiple times over two years — first iteration predated modern tool-calling, last iteration rebuilt the PowerPoint subagent on top of the claude-code SDK. On hiring, he's updated his bar: a year ago he'd hire smart engineers with no AI experience; by 2026 he considers it "a bit strange if you've never worked with AI" after three and a half years of the wave.
Believes in end-to-end ownership: the engineer who builds an agent is the same person reading its traces, tuning its prompts, and owning production regressions — no separate prompt-writer, product-manager, or eval-labeler role. His pushback on letting non-engineers "just add a sentence to the prompt" is explicit: "what about the validation? Who's actually going to take the blame?"
Key ideas he's carrying¶
- virtual-table-architecture over virtual file systems for tabular qualitative data
- contextual-prompt-engineering as an alternative to skills
- feedback-subagent as both an async-run reviewer and a live-run eval signal
- trigger-long-running-agent — decide sync-chat vs async-run from the first prompt