You've built agents that supercharge your development workflow. They're wired into your tools through MCPs, they handle the grunt work, and your team is shipping faster. Now you're seeing agent potential everywhere: internal tooling, operations, customer-facing features. But scaling beyond IDE customization requires a different skillset: building agents as operational software systems.

This talk maps the evolution from IDE agents to internal productivity agents to customer-facing systems, with deep focus on the first critical transition. The core challenge? IDE agents are configuration and local workflows. Internal productivity agents are production software engineering with cloud resources, deployed code, agent frameworks, monitoring, and disciplined prompt and eval control. Most teams know how to customize VS Code but haven't architected autonomous systems that run unsupervised.

By treating internal productivity as a controlled-risk learning environment, you build operational trust through dogfooding before the stakes involve customers. Using real examples from one product context, we'll explore the architecture and tooling patterns that distinguish operational agent systems from IDE customization. Then we'll look toward customer-facing agents as the horizon: what skills from internal productivity transfer forward, and how to know when you're ready.

You'll walk away knowing how to identify where you are today, the operational patterns needed to transition from IDE customization to production agent systems, and what readiness signals indicate you can move toward customer-facing work.