At Snyk, we built the world's first agentic orchestrator for security. We didn't want to just put chat into an existing product, so we focused on solving real customer problems. Problems like seeing where your AI, agents and MCPs are; threat modeling and red teaming AI native applications; understanding supply chain risks in the AI world; personalised reporting and leveraging multi-agent architectures.

It took every trick in the book, across UX, engineering, and AI to deliver. How do we deliver a chat based experience, but not require customers to wait minutes for the application to respond? How do we secure the agent against prompt injections? How do we steer the agent to pick the right tool and perform the same action?

Key takeaways include:

• Striking a balance between chat based and click based UX
• Securing agents in production
• Building integrations and tools using MCP