Microsoft has aimed the spotlight on generative AI innovation during this year’s Build developer conference. Highlighting this focus, the company has rolled out enhancements to its suite designed for creating AI-infused applications, namely Azure AI Studio and Copilot Studio.
Azure AI Studio equips users with a rich set of tools as part of the Azure OpenAI Service. Here, developers have the ability to integrate models such as OpenAI’s advanced GPT-4 with personalized datasets. This integration enables the creation of dynamic chat assistants and similar applications that can intelligently interact with the user’s data. Copilot Studio, on the other hand, offers developers resources to link the AI capabilities within Microsoft 365’s Copilot—including in apps like Excel, Word, and PowerPoint—with external data sources.
Now broadly accessible, Azure AI Studio is introducing a cost-effective, pay-as-you-go approach for developers to harness generative AI power through inference APIs. Initially featuring models from Nixtla and Core42, it plans to expand with offerings from Cohere, Stability AI, and AI21 Labs. Dubbing it “model-as-a-service,” Microsoft is paving the way for simplified access to hosted AI model functionalities.
Among the additional features, Azure AI Studio now in preview, allows for in-depth app training and debugging. Developers can compare app versions and actively monitor their performance upon deployment, setting custom alerts and filters to stay informed on app usage and quality metrics. Additionally, an integration with Microsoft Purview ensures that developers can identify potential data risks in their AI applications, apply encryption to sensitive content, and efficiently govern app usage.
In an effort to reinforce security, Azure AI Studio also introduces tools to curtail AI model “jailbreaks,” which are tactics that bypass built-in model safeguards. Furthermore, it enhances the ability to detect “hallucinations,” instances where the model fabricates information.
Turning the page to Copilot Studio, Microsoft has unveiled the advent of Copilot agents—intelligent AI bots that are able to autonomously manage tasks custom-tailored to specific roles and functions. These agents are designed to learn from user interactions and to seek assistance when they meet unknown challenges. In essence, they can navigate complex business processes, optimize workflows, and facilitate task automation discreetly in the background.
More innovations in Copilot Studio include the introduction of extensions and connectors, both in a preview phase that extends to Microsoft 365 and the Teams enterprise collaboration environment. Extensions enable developers to finetune Copilot’s AI to perform specific tasks—such as automated expense reporting and employee onboarding—by feeding it tailored instructions and relevant data from databases or plugins. Connectors serve as vessels that provide the AI with access to a wealth of organizational knowledge from various data sources.
Extensions and connectors offer a robust array of functionalities, from augmenting Copilot actions and grounding it with pertinent business data to enabling hand-offs to other AI systems. These expansions embrace a comprehensive range of data sources, covering everything from public web content to SharePoint, OneDrive, Dataverse tables, and leading third-party applications.
As AI technology continues to evolve at a rapid pace, Microsoft’s latest enhancements to its AI app-building platforms advance the capabilities and accessibility of generative AI applications. With these tools, developers can create more intelligent, responsive, and secure apps, forged to handle modern-day business complexities and data governance requirements.
