At the Innovation Center, we keep a close eye on the latest advances in artificial intelligence and other emerging technologies. Our goal goes beyond simple awareness of these developments; we seek to understand the technical details of how they work. We regularly experiment with new tools, build our own prototypes, and explore practical applications so we can share meaningful insights with the university community. One concept that has been receiving increasing attention in the AI world is “Agentic AI.” Unlike traditional AI systems like ChatGPT that simply respond to prompts, agentic AI systems can plan and carry out sequences of tasks in order to accomplish a goal. For example, an agentic system might receive a request to “research a topic, summarize the results, and save them to a document,” then automatically perform each step—searching the web, analyzing the information, and generating a summary.

Of course, with systems that can take actions on a user’s behalf, issues such as privacy, data integrity, transparency, and general system safety become critically important. For that reason, we have begun a series of internal projects to build our own in-house agentic AI system with safety and transparency as core design principles. This effort builds on several earlier Innovation Center research mini-projects involving speech-to-text, large language model processing, and text-to-speech technologies. Our current architecture works by taking a user request and having a large language model act as a “planner” to convert the request into a structured set of tasks. That plan is then analyzed by a “validator” to ensure it aligns with system capabilities and safety rules. For instance, some actions such as deleting files may be disallowed, while others—like web searches—require specific information before they can proceed. Once validated, the task list is passed to an “executor” component that carries out the approved actions and reports the results.

These tasks can include practical activities such as opening software on a computer, dictating notes and saving them to a specific file and folder, or performing a web search and summarizing the top results. By designing and building these systems ourselves, we gain a deeper understanding of how they function, what risks they present, and how they can be used responsibly. As always, the Innovation Center remains committed to supporting the university community by exploring new technologies and sharing the knowledge, tools, and insights that can help students, faculty, and staff work more effectively in their day-to-day activities.