An Exploration into The Capabilities of Microsoft Copilot and Its AI Agent Creation

Abstract

Microsoft Copilot's Agent Creation process was explored and investigated for the public sector municipality Hämeenlinna City. The proposed task was to create three agents designed for a specific purpose in order to explore and test the capabilities of Copilot and its agents. The three agents created were a Public Announcement Generation Agent, a Speech Generation Agent and a Data Analysis Agent. The frameworks utilized were the RTF Request-Task-Framework and the Microsoft Agent Creation framework. Observations and issues were noted during the creation and testing; these were reported in order to generate recommendations on usage and implementation of Copilot agents within the City departments.

The Copilot Agent Creation process is straightforward and simple, created with the aid of an internal AI helper, therefore any employee with access could take advantage of the software to improve their day-to-day efficiency as long as the agent's purpose was targeted, specific and simple. Main recommendations from observations and issues were “one conversation, one topic”, new conversations with agents must be generated for each request. Large files caused issues so reduction in size or focus on the area to be analyzed, additionally the wrong chart or diagram was often generated. Agents handle refined and focused specific requests better than wider general requests.