How Hex builds AI agents that reason like human data analysts | Izzy Miller, AI Engineer

<p>Izzy Miller is an AI engineer at Hex, an AI analytics platform that was one of the first companies to ship data agents to real paying users. Today, Hex runs a multi-agent system with nearly 100K tokens of tools, and Izzy is building a 90-day simulation to evaluate whether those agents actually get smarter over time. In this conversation, he walks through the harness decisions that shaped their architecture, the failure modes Hex is seeing at scale, and what it takes to build an eval that no current model can pass.</p><p><br></p><p><strong>We also discuss:</strong></p><ul><li>Why data agents are harder to verify than coding agents</li><li>Under the hood of Hex’s agents</li><li>How Hex is unifying separate agents</li><li>Why most eval sets are bad</li><li>The 90-day simulation for long-horizon evals</li><li>How Izzy went from marketing to AI engineer</li></ul><p><br></p><p><strong>References:</strong></p><ul><li><a href="https://andonlabs.com/" target="_blank" rel="ugc noopener noreferrer">Andon Labs</a></li><li><a href="https://www.anthropic.com/" target="_blank" rel="ugc noopener noreferrer">Anthropic</a></li><li><a href="linkedin.com/in/barrymccardel" target="_blank" rel="ugc noopener noreferrer">Barry McCardel</a></li><li><a href="http://chatgpt.com" target="_blank" rel="ugc noopener noreferrer">ChatGPT</a></li><li><a href="https://code.claude.com/docs/en/overview" target="_blank" rel="ugc noopener noreferrer">Claude Code</a></li><li><a href="https://www.anthropic.com/news/claude-sonnet-4-6" target="_blank" rel="ugc noopener noreferrer">Claude Sonnet 4.6</a></li><li><a href="https://www.getdbt.com/" target="_blank" rel="ugc noopener noreferrer">DBT</a></li><li><a href="https://developers.openai.com/api/docs/models/gpt-3.5-turbo" target="_blank" rel="ugc noopener noreferrer">GPT-3.5 Turbo</a></li><li><a href="https://openai.com/index/introducing-gpt-5-3-codex-spark/" target="_blank" rel="ugc noopener noreferrer">GPT-5.3 Codex Spark</a></li><li><a href="https://openai.com/index/introducing-gpt-5-4/" target="_blank" rel="ugc noopener noreferrer">GPT-5.4</a></li><li><a href="https://hex.tech/" target="_blank" rel="ugc noopener noreferrer">Hex</a></li><li><a href="https://www.langchain.com/" target="_blank" rel="ugc noopener noreferrer">LangChain</a></li><li><a href="https://www.smith.langchain.com/" target="_blank" rel="ugc noopener noreferrer">LangSmith</a></li><li><a href="https://lookerstudio.google.com/" target="_blank" rel="ugc noopener noreferrer">Looker</a></li><li><a href="https://openai.com/" target="_blank" rel="ugc noopener noreferrer">OpenAI</a></li><li><a href="https://www.anthropic.com/news/claude-opus-4-6" target="_blank" rel="ugc noopener noreferrer">Opus 4.6</a></li><li><a href="https://www.linkedin.com/in/satyanadella" target="_blank" rel="ugc noopener noreferrer">Satya Nadella</a></li><li><a href="https://www.snowflake.com/en/" target="_blank" rel="ugc noopener noreferrer">Snowflake</a></li><li><a href="https://andonlabs.com/vending" target="_blank" rel="ugc noopener noreferrer">Vending Machine</a></li></ul><p><br></p><p><strong>Where to find Izzy:</strong></p><ul><li><a href="https://www.linkedin.com/in/izzy-miller/" target="_blank" rel="ugc noopener noreferrer">LinkedIn</a></li><li><a href="https://x.com/isidoremiller" target="_blank" rel="ugc noopener noreferrer">Twitter/X</a></li></ul><p><br></p><p><strong>Where to find Harrison:</strong></p><ul><li><a href="https://www.linkedin.com/in/harrison-chase-961287118/" target="_blank" rel="ugc noopener noreferrer">LinkedIn</a></li><li><a href="https://x.com/hwchase17" target="_blank" rel="ugc noopener noreferrer">Twitter/X</a></li></ul><p><br></p><p><strong>Where to find LangChain:</strong></p><ul><li><a href="http://langchain.com" target="_blank" rel="ugc noopener noreferrer">Website</a></li><li><a href="https://docs.langchain.com/" target="_blank" rel="ugc noopener noreferrer">Docs</a></li></ul><p><br></p><p>Send feedback or questions to maxagency@langchain.