The Data Exchange with Ben Lorica How to Build and Optimize AI Research Agents
19 snips
Sep 25, 2025 Jakub Zavrel, CEO of Zeta Alpha and expert in AI research agents, dives into the evolution from traditional enterprise search to advanced deep research systems. He contrasts user expectations of fast web searches with the thorough, report-style answers of deep research. Zavrel discusses the importance of customization and quality optimization, and his innovative JEPA method for prompt evolution that outperforms traditional reinforcement learning. He also emphasizes the need for secure enterprise search as a foundation for powerful AI solutions.
AI Snips
Chapters
Transcript
Episode notes
Deep Research vs Traditional Search
- Deep research produces comprehensive reports rather than quick summaries to answer complex knowledge-intensive questions.
- It compresses hours or days of research into minutes by iterating search and LLM reasoning.
Completeness Over Speed
- Deep research trades raw speed for completeness and uses agentic AI to run multi-step searches and assemble results.
- Users accept slower workflows if the outcome is a trustworthy, high-value report.
Build Modular Agent Ensembles
- Use modular ensembles of agents with specialized prompts to gain control over report direction and format.
- Optimize each agent separately when you cannot fine-tune large models at scale.