Wen Sang

EP 73 Genspark's Bet: AI Agents Become the Users of Software | Wen Sang

• Genspark builds the vehicle around the frontier labs’ engines. Wen Sang describes a “mixture of agents” execution layer that routes work across 70+ models by strength (reasoning models plan the project, coding models build a presentation in HTML first, multimodal models handle image and video, fast and cost-effective models take lighter jobs like visualizing data), backed by 150+ in-house tools and paid data sets such as PitchBook, Crunchbase and S&P. Episode takeaways
• Genspark’s evals grade the work, not the model. Sang says the frontier labs focus on making models smarter, while Genspark’s recursive learning system grades every interaction on whether the user got boardroom-ready output and on whether the result is reliable, accurate and usable. His own example: whether a sales proposal responded to every point in the RFP. Episode takeaways
• A meeting note is the start of the work, not the deliverable. A SecondBrain Note recording can be turned into a sales proposal, a pricing model, an ROI analysis, a client presentation, or research on the open questions, and the note is folded into a memory layer Sang calls the second brain so the agent already knows the user’s priorities and relationships. Episode takeaways