Microsoft’s emphasis at this year’s Build conference is firmly on generative AI, heralding a slew of enhancements to its platforms designed for creating AI-powered applications and experiences: Azure AI Studio and Copilot Studio.
To recapitulate, Azure AI Studio and Copilot Studio are integral components of Microsoft’s AI ecosystem. Azure AI Studio, part of the Azure OpenAI Service, enables customers to integrate an AI model such as OpenAI’s GPT-4 with their own datasets to develop chat assistants or other applications that can intelligently analyze that data. Conversely, Copilot Studio offers tools to link Copilot for Microsoft 365—which powers AI-driven features in applications like Excel, Word, PowerPoint, and in the Edge browser and Windows OS—to external data sources.
Now generally available, Azure AI Studio soon equips developers with the ability to create generative AI applications through pay-as-you-go inference APIs. These APIs provide access to and customization of generative AI models hosted on Azure’s robust infrastructure. This approach, termed “model-as-a-service” by Microsoft, initially features models from Nixtla and Core42, with future expansion plans to include models from other providers such as Cohere, Stability AI, and AI21 Labs.
Additional preview features in Azure AI Studio include capabilities for training and debugging generative AI applications by comparing their different versions, as well as monitoring them in production to assess usage and quality. Users can identify trends and set up custom alerts based on specific filters and criteria.
Moreover, Azure AI Studio now integrates with Microsoft Purview (currently in preview) to mitigate unauthorized data access across various applications and services. This integration facilitates the identification of potential data risks in AI applications, enforces encryption on sensitive data, and governs AI app usage. New tools have also been introduced to thwart “jailbreaks”—techniques that disable an AI model’s safety mechanisms—and to detect and address instances where an AI model fabricates information.
On the Copilot Studio front, Microsoft is introducing Copilot agents, AI-driven bots designed to “independently orchestrate tasks tailored to specific roles and functions,” as per the company’s description. These agents utilize memory and contextual knowledge to navigate diverse business workflows, learning from user interactions and seeking assistance when faced with unfamiliar situations.
Charles Lamanna, Corporate Vice President of Business Applications and Platforms at Microsoft, elaborates on the concept in a press release: “Developers provide their copilot with a specific task, equip it with the requisite knowledge and actions, and then Copilot Studio manages dynamic workflows and operates behind the scenes to integrate them and automate the task.”
Additionally, Copilot Studio is unveiling extensions and connectors, currently in preview for Copilot for Microsoft 365 and incorporated within Teams, Microsoft’s enterprise collaboration platform. Extensions enable developers to personalize AI copilots with tailored instructions, insights from databases, and actions from plugins, thus creating copilots capable of managing tasks like expense reporting and employee onboarding. Conversely, connectors offer developers methods to “ground” a copilot in organizational knowledge sourced from various channels.
According to Lamanna, extensions broaden the scope of actions that Microsoft Copilot can perform on behalf of users. They also tailor the grounding knowledge by incorporating pertinent business data and facilitate the transition to other copilots. Additionally, Copilot connectors encompass Power Platform connectors, Microsoft Graph connectors, and Power Query connectors, with upcoming integrations for Microsoft Fabric. This integration allows copilots to access a variety of data sources, including public websites, SharePoint, OneDrive, Dataverse tables, OneLake by Microsoft Fabric, and Microsoft Graph, as well as major third-party applications.
