May 2026 Summaries

Exa, a modern AI search engine, has raised $250 million in a Series C funding round led by a16z, valuing the company at $2.2 billion. The company aims to revolutionize information retrieval for AI agents, which are projected to surpass human web searches this year. Exa's offerings include a web search API designed specifically for AI products, which has attracted over 5,000 companies, such as HubSpot and Monday.com, utilizing its services. Exa focuses on various verticals like coding and GTM agents by fine-tuning embedding models and building comprehensive indexes, promising a significant improvement in search quality, latency, and cost compared to competitors who wrap other search engines. With its recent funding, Exa plans to enhance its infrastructure and expand its workforce by hiring top talent from companies like Meta and Google to meet the anticipated demand from AI agents, which require precise, comprehensive, and fresh information at an unprecedented scale. The company believes that achieving perfect search is crucial for both AI and human access to high-quality information, especially in an era characterized by political fragmentation and rapid technological advancements.

The study explores the impact of different search backends on reinforcement learning (RL) outcomes by comparing an agent trained with Exa against one trained with a SERP-based backend. The research finds that agents trained with Exa outperform those trained with SERP in terms of pass@k performance across various benchmarks, achieving higher accuracy with less computational cost in both training and inference phases. The Exa-trained agents demonstrated better sample efficiency, retrieving more relevant information with fewer actions, which enhanced learning and reduced the sparsity of rewards. Furthermore, the Exa-trained agents maintained superior performance even when the search backend was switched at inference, suggesting that the skills learned with Exa are transferable. This indicates that the choice of search engine significantly affects the efficiency and effectiveness of RL training for language models, emphasizing the importance of using a robust search backend like Exa for optimal results.