# Your Coding Agent Is 6 Months Out of Date — Jakub Hojsan, Exa **URL:** \ **Category:** Content Creators\ **Tags:** agents, ai-engineer, search, coding\ **Created:** [2 October 2026 15:30 UTC](https://www.artofsm.art/t/your-coding-agent-is-6-months-out-of-date-jakub-hojsan-exa/24880 "2026-10-02T15:30:31Z")\ **Posts on this page:** 2\ **Page:** 1
**Author:** ![artesia](https://www.artofsm.art/user_avatar/www.artofsm.art/artesia/32/36_2.png) [@artesia](https://www.artofsm.art/u/artesia)\ **Post date:** [2 October 2026 15:30 UTC](https://www.artofsm.art/t/your-coding-agent-is-6-months-out-of-date-jakub-hojsan-exa/24880/1 "2026-10-02T15:30:31Z")
[![](https://www.artofsm.art/uploads/default/original/3X/9/9/99a10aef852832d986cb84a5533bf1218f456367.jpeg "Your Coding Agent Is 6 Months Out of Date — Jakub Hojsan, Exa") ](https://www.youtube.com/watch?v=cKhpeEBnT1o) Jakub Hojsan from Exa highlighted the limitations of large language models due to their six-month knowledge lag and introduced Exa’s semantic search engine that delivers concise, context-rich snippets to keep coding agents up-to-date without overwhelming them with data. He also presented Exa Agent, a turnkey solution that orchestrates specialized search experiences across industries, offering transparency, token efficiency, and multi-provider support to enhance coding reviews and other domain-specific applications. ---
**Author:** ![artesia](https://www.artofsm.art/user_avatar/www.artofsm.art/artesia/32/36_2.png) [@artesia](https://www.artofsm.art/u/artesia)\ **Post date:** [2 October 2026 15:54 UTC](https://www.artofsm.art/t/your-coding-agent-is-6-months-out-of-date-jakub-hojsan-exa/24880/3 "2026-10-02T15:54:44Z")
Jakub Hojsan, a Ford deployed engineer at Exa, presented on building search capabilities for coding and review agents, emphasizing the limitations of large language models (LLMs) due to their knowledge cutoffs, which typically lag about six months behind real-time developments. This gap creates challenges in reviewing recent code changes or pull requests that fall outside the model’s training data. To address this, Exa developed a semantic search engine that provides context-rich, distilled highlights from relevant sources, enabling LLMs to access up-to-date information without overwhelming them with excessive data. Hojsan illustrated the problem with an example involving a dependency bump in a GitHub repository, where a human reviewer might turn to Stack Overflow to understand compilation errors or code changes. Unlike traditional web search results, Exa’s approach returns concise, relevant snippets rather than entire pages, allowing the LLM to efficiently interpret and explain code changes. This method avoids feeding the model large volumes of unnecessary text, instead delivering only the essential 500 characters needed to answer specific queries. Adding a web search tool to an agent is not sufficient on its own; it requires instructing the agent on when and how to use the search effectively. Hojsan explained that simply integrating a search API often fails because the model may not recognize new versions or dependencies without explicit rules guiding its use. Exa’s solution involves a two-step process: instructing the agent on when to perform searches and then executing those searches to ground the agent’s responses in current, verified information. Exa’s search platform offers several advantages over native web search tools integrated into LLM providers. It provides full transparency by exposing the exact queries, sources, and highlights passed to the model, enabling better debugging and telemetry. The platform is also token-efficient, reducing costs by distilling large documents into concise highlights. Additionally, Exa supports multiple model providers through a standardized API, offering flexibility and avoiding vendor lock-in. This approach has been adopted by several prominent coding tool companies and is expanding into other domains like financial analytics and podcast intelligence. Finally, Hojsan introduced Exa Agent, a new offering that orchestrates search experiences without requiring users to build their own search infrastructure. By partnering with various data providers, Exa Agent delivers specialized insights across industries, including finance and events. Hojsan concluded by inviting attendees to try Exa’s services, highlighting its broad applicability and ease of integration, and answered questions about document reranking and schema generation, demonstrating the platform’s advanced capabilities and user-friendly features. ## Useful Links - [Exa Official Website](https://exa.ai/) — Central to the video as it is the platform providing the search solution for coding agents. - [Exa MCP Documentation](https://docs.exa.ai/reference/exa-mcp) — Directly explains how to use the Exa search platform described in the video. - [Exa MCP Server GitHub Repository](https://github.com/exa-labs/exa-mcp-server) — Provides source code and implementation details for the Exa search platform discussed in the video.