Paul Henry is the founder and chief executive of BotDojo, which helps companies build, evaluate, and operate AI agents. Before founding the company, he was chief technology officer of Auctane, overseeing engineering for a shipping-software business serving more than a million customers and processing more than three billion shipments annually.
Henry joined ShipStation when it had roughly ten employees and remained after its acquisition by Stamps.com in 2014. Over approximately a decade, he helped expand the business through new products, partnerships, and acquisitions, eventually leading Auctane’s global engineering organization. His account of departing Auctane also names engineering leader Edwin Woo, now BotDojo’s chief technology officer.
At AI Engineer World’s Fair 2024, Henry demonstrated a support assistant that traced questions through conversation history, vector retrieval, model execution, and structured responses. Batch evaluations exposed missing documentation and potential hallucinations even when answers appeared polished and included citations.
- Evaluation-driven AI deployment: Henry uses historical customer questions to test prompts, retrieval strategies, and models before release, while node-level traces reveal whether failures originate in missing context, unreliable tools, or the model itself.
- Synthetic documentation grounded in support conversations: His workflow converts real customer questions into test cases, drafts documentation from existing indexed material, checks whether it answers the original question, and adds useful content to the knowledge base. Repeating the evaluation measures whether the assistant improved.
- Context and governance before complexity: His deployment guidance emphasizes relevant business information, straightforward instructions, practical tools, observability, spending limits, execution safeguards, and user-specific permissions.
- Interface-aware AI agents: Henry has explored multi-agent tool calling across models and advanced BotDojo’s Interactive Agent SDK, which uses the Model Context Protocol to connect assistants with visible interface context, form fields, controls, and embedded interactive elements.
