Ruben Casas

Ruben Casas is a Postman staff engineer and Google Developer Expert developing secure generative interfaces for AI agents. His career spans frontend infrastructure, developer tools and application performance, with a growing focus on how people and agents can safely collaborate inside interactive software.

Originally from Colombia and based in the United Kingdom, Casas studied internet and distributed systems at the University of Brighton. At American Express, he helped maintain One App, an open-source framework for composing independently developed, server-rendered React applications. His writing on micro-frontends and React Router emphasizes nested routing, explicit data ownership and gradual migration: architectural choices that give teams independence without creating a distributed monolith.

At Postman, Casas helped ship the company’s Visual Studio Code extension and improve the startup performance of its Electron application. His account of accelerating Postman’s launch highlights user-centered measurement, bottleneck analysis and coordination across engineering teams. He also advocates Time to Usability as a more meaningful benchmark than whether an interface merely appears loaded.

His work subsequently expanded into agent-facing developer tools, including MCP-UI support for Postman’s MCP Inspector and Agent Mode, testing capabilities for ChatGPT Apps, and MCP Apps integration. His priorities include:

• Isolating untrusted agent-generated applications. Casas implemented a double-iframe security architecture that separates embedded applications from their host through distinct origins and controlled messaging. He also reverse-engineered the ChatGPT Apps sandbox to make those applications inspectable and debuggable in Postman. He treats model-generated code with the same caution as unfamiliar third-party software.

• Matching interface architecture to actual risk. In his analysis of agent-driven interfaces, Casas distinguishes host-controlled components, third-party MCP applications and fully generated interfaces. He considers declarative generative UI a practical intermediate approach: models produce structured JSON or YAML while trusted components preserve design consistency, predictable behavior and lower generation costs.

• Making agents participants in shared software. Casas built a weather-agent experiment that combined an external API, generated humor and dynamically created HTML, CSS and JavaScript within one tool call. His AI Engineer Europe session argues that human-agent collaboration should extend beyond chat windows toward editable canvases, shared artifacts and interfaces that both people and agents can change. MCP Apps provide authentication, tool calling, message passing and sandboxing that make those interactions more practical to distribute safely.