AI BriefingHarmonic Rebuilds Scout with Deep Agents and LangSmith, Achieving 4x Retention
Key point
Harmonic rebuilt Scout using Deep Agents and LangSmith, increasing user retention by 4x.
Details
Starting as a venture capital sourcing tool, Harmonic enabled investors to query private market data in natural language through its AI interface Scout. The initial version, based on LangGraph's static subgraph structure, required hundreds of evaluations and complex tuning for every new feature, resulting in high maintenance costs.
Lead engineer Austin Berke rebuilt Scout V2 on a Deep Agents harness, the exact opposite of the previous approach. The new architecture adopts a simple design that connects a single frontier model to Harmonic's global data layer (40 million companies, 200 million people, 230,000 investors) and tools for querying entity-specific context.
This change shortened the product iteration cycle from months to days, allowing users to receive reliable advice at the level of an 'investment advisor' rather than just using a simple search tool. With production operational stability secured through LangSmith Deployment, expansion into a larger Total Addressable Market (TAM) became possible.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.