Descriptions:
João Moura, CEO of CrewAI, delivers a fast-paced session at AI Dev SF 2026 that weaves together CrewAI’s own internal experience deploying agents with patterns emerging across their enterprise customer base. The anchor case study is Iris, an internal AI coding agent built on CrewAI that went from a skeptical reception (engineers immediately tried to break it) to handling roughly half of all pull requests at the company in a matter of months. Notably, Iris now writes its own skills, maintains its own memory, and updates itself—an example of a self-improving production agent rather than a static automation.
Moura identifies three converging dynamics shaping enterprise agent adoption: ad hoc agents (one-off, interactive use) blurring into embedded workflows (recurring, governed automations); building becoming commoditized as AI coding tools proliferate; and the rising importance of reusable building blocks—tools, agents, and integrations—that can be consumed by both humans and other agents. CrewAI’s response includes native support for MCP, A2A, LangGraph, ADK, and Salesforce agents within a single orchestration layer, and a Claude Code integration that reduced friction for developers who prefer terminal-based workflows.
The session is particularly useful for engineering and product leaders thinking through how to move from pilot agents to governed, recurring enterprise workflows—and for teams evaluating multi-agent orchestration frameworks that interoperate with a heterogeneous mix of agent runtimes.
📺 Source: DeepLearningAI · Published May 22, 2026
🏷️ Format: Workflow Case Study
