<p>LLM-based agents are everywhere, but most research focuses on just one step: getting the model to call the right tool. What happens after that?</p><p><br></p><p>Paper: <a href="https://arxiv.org/abs/2510.15955" target="_blank" rel="noopener noreferer">https://arxiv.org/abs/2510.15955</a></p><p><br></p><p>In this talk, Kiran Kate (IBM Research) presents new findings from their EACL 2026 paper on a largely overlooked problem:👉 Can LLMs actually understand and use the outputs returned by tools?</p><p><br></p><p>As tool-augmented systems become more complex, this question becomes critical. The work dives into how current models handle non-trivial, real-world tool responses, and where they break down.</p><p><br></p><p>💡 Key ideas covered:</p><ul><li>Why tool calling is only half the story in LLM agents</li><li>The challenge of processing complex tool outputs</li><li>Failure modes in current LLM-based systems</li><li>What this means for building robust, real-world AI agents</li></ul><p><br></p><p>This talk is especially relevant if you&#39;re working on:</p><ul><li>LLM agents and tool use</li><li>Evaluation of LLM capabilities</li><li>Real-world deployment of AI systems</li><li>Agentic workflows and reasoning pipelines</li></ul><p><br></p><p>Kiran Kate: </p><ul><li><a href="https://www.linkedin.com/in/kiran-kate-8b98672/" target="_blank" rel="noopener noreferer">https://www.linkedin.com/in/kiran-kate-8b98672/</a></li></ul><p><br></p><p>👍 Like &amp; subscribe for more deep dives into cutting-edge AI research</p><p>🔔 New episodes from EACL 2026 coming soon</p>