A Model Plus an API Is Not a Product - Danilo Delizia, commercetools

How do you give an agent hundreds of tools? Danilo Delizia, staff software engineer at commercetools, shares how his team built an agent harness on Mastra that lets business teams run their commerce operations just by asking. commercetools is an API-first commerce platform with hundreds of endpoints across dozens of domains, and enterprise customers connect tools of their own. Most agent frameworks expect you to define your tools up front. Here they change with each merchant and with what each user is allowed to do, so the team wrote its own agent loop. You'll learn how they keep it accurate as the tool catalog grows. Skills give the agent a playbook for each domain and load only when a question needs them. Tool search lets the agent find the right MCP tool at runtime, so it never has to see the whole catalog. MCP annotations mark which tools only read data and which change it, and a human approves every change. Big results go into DuckDB, where the agent queries them with SQL, so a question about 3,000 orders gets answered from five rows. If you expose every tool at once, they might not fit in the context window. And when they do fit, the model can start to hallucinate. He also shares the story of a revenue total that looked right and wasn't, and why the team traces every step the agent takes. Recorded at the TypeScript AI Conference in London, July 23, 2026, presented by Mastra. 🔗 MENTIONED IN THIS TALK commercetools: https://commercetools.com DuckDB: https://duckdb.org 📚 MASTRA RESOURCES https://mastra.ai https://x.com/mastra_ai https://mastra.ai/community/discord https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn https://mastra.ai/books/principles-of-building-ai-agents https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development — from prototype to production. CHAPTERS 00:00 Intro 00:32 Danilo Delizia and commercetools 01:23 Why build our own harness 02:13 No fixed set of tools 02:58 Three decisions 03:33 Lesson one: tool surface area 05:08 Skills and tool search 06:05 Read tools and write tools 06:48 Lesson two: context 07:46 One question, end to end 08:54 Lesson three: observability 09:35 Wrap-up