Azure AI Foundry Unleashes a New Era of Intelligent Agents

REDMOND, ​Washington, March 31, 2025 – In a year already full of artificial intelligence, Microsoft has pushed the conversation back with significant improvements to its Azure AI foundry platform. ​These innovations not only push the platform, but redefine what AI agents can ​do for modern businesses. At the heart of this transformation is a fusion ​of next-generation data orchestration, smart automation and deep collaboration between Microsoft and NVIDIA that indicates a seismic change in how organizations will deploy and manage AI systems on ​a scale.

Since the transformation of global business tasks to support highly customized generic IA applications, Azure AI Foundry now functions as an agent factory and strategic ​nerve center. According to Microsoft, these improvements will enable ​companies to implement AI more quickly, safely and cost-effectively: bringing them closer to the difficult goal of scalable and smart automation. But what exactly has ​changed, ​and what does ​that ​mean for the deployment of business AIs?

What is the Azure ​AI foundry, and why is it important now?

Launched in late 2024, Azure ​AI Fundry was introduced ​as a ​unified framework for the ​construction, training, deployment and ​management of AI agents. Think of it as a digital intelligence plant, with pre-built components, modular ​workflows and operational feedback loops. More than 60,000 customers have adopted the platform to date, using it to build agents that manage everything, from RDH internal requests ​to the optimization of customer services, according to Microsoft reports.

But the relevance of the platform has ​increased considerably in a few months. While AI goes beyond static models in dynamic ​and multi-agent systems, companies struggle to ensure consistent performance, secure integration ​and fast customization. Azure AI Foundry strives to address these concerns, not only with tools, but with a philosophy: One that suits your needs, not the other way around. This adaptability is essential for industries that try to remain agile in the context of data, regulation and client expectations.

What new features are available ​now?

The latest series of Microsoft innovations includes several tools and agent capabilities aimed at putting intelligent systems ​to work with minimal friction. A confrontation between them is the Azure AI Agents ​Service, a framework for the creation of AI agents based on secure business knowledge. These are ​not just models of large pre-formed languages (LLM); The aim ​is ​to ensure that these models understand, ​have access to private, ​structured or informal ​data and that they are driven by sources of trust.

In addition, Microsoft Fabric data agents were introduced, allowing developers to create ​conversation agents ​that are inserted into structured ​semantic data. These agents are able to access and interpret Microsoft OneLake business data, deciding when and how to combine multiple ​data sets. It’s like giving AI ​the brain energy of ​a data analyst with the memory ​of a librarian, in real time.

How do companies already use these tools?

NTT ​DATA is already studying the impact ​of these capabilities on the real world. According to Microsoft, the company deployed data agents connected to the fabric ​to interact with HR and back-office systems. These IA officers can synthesize and analyze ideas from previously silozed data sets, helping managers understand ​internal operations in a more nuanced and dynamic way.

Another example comes from Atomicwork, including AI Agent Atom, which helps simplify IT and HR workflows. Users reported significant improvements in ​operational efficiency, employee satisfaction and cost reduction. Atom ​not only responds to ​commands, but proactively solves problems, makes ​suggestions and learns from each ​interaction.

What is the role of ​the new computer agent (CUA)?

Among the most intriguing announcements is the launch of the Computer Usage Agent (ACU). ​This model goes beyond text generation or data retrieval. simulates human-computer interactions, browser software interfaces, pressure buttons, filling forms and automation of end-to-end workflows. The AUC is essentially a ​digital employee, able to integrate disparate systems without ​relying on APIs.

Imagine a ​AUC processing customer service from ​an email input tray, connect to CRM software, input and tracking data, all independently. This evolution of automation saves time and eliminates the need for tedious custom integrations. It also highlights the ​shift from AI passive auxiliaries to active ​partners in ​digital equipment capable of transforming work dynamics.

What performance improvements have been announced?

To improve scale performance, Microsoft has been closely associated with NVIDIA, integrating NVIDIA NIM microservices into Azure AI Foundry. ​As Microsoft has said, these microservices allow high performance IA inference with off-box optimization and ​perfect integration with Azure IT infrastructure.

The main features are:

• Zero-configuration deployment: Start running high-efficiency inference with minimal ​setup.
• Azure-native integration: Full compatibility with Semantic Kernel and AI Agent Service.
• Enterprise-grade ​reliability: ​Backed by NVIDIA’s AI Enterprise suite for continuous optimization.

According to Microsoft, ​these capabilities allow companies to maximize the use of GPU, reduce cost overload and maintain high performance under ​pressure, a critical factor in the deployment of AI at the company level.

How ​does ​Microsoft ensure that the IA is responsible?

One of the most pressing concerns with advanced IA officers is ​trust. Can these officers be trusted ​to behave responsibly on a scale? To remedy this, Microsoft introduced AI Red Teaming Agent, now in public ​view. As Microsoft reported, this tool systematically tests AI models ​for vulnerabilities using Microsoft ​Security’s PyRIT framework, followed by improvements over time.

In addition, there is a new set of assessment measures for officers who offer risk and quality assessments. These ​tools not only monitor performance, but help developers identify and mitigate potential threats at the beginning of ​the development cycle. Organizations such as Accenture are already exploring these security tools to ensure that their requests for ​agents meet the standards of compliance and governance.

What about ​the developers? ​Is it easier to build now?

Sure. Microsoft introduced a ​new ​Azure AI Foundry extension for Visual Studio Code, facilitating ​developers to ​build, test and implement agents directly in their IDE. The Agent ​mode for GitHub Copilot has also received ​a major improvement, allowing developers to perform ​complex multi-step operations, automated code reviews and tests ​without leaving their environment.

According to ​Microsoft, these tools reduce context change and ​help developers move faster while ​maintaining code quality. ​This ​is another step to make ​AI development intuitive and not intimidating.

How do Microsoft and NVIDIA ​collaborate?

The strategic partnership between Microsoft and NVIDIA was further developed through its joint work with ServiceNow. As announced, ​the trio is working on advanced tools for assessing and integrating AI ​agents with the Flame Nemotron models of NVIDIA, which provide more reasonable and adaptable AI agents.

The new ServicNow tools allow companies to assess the performance of IA agents in key areas such as accuracy and transparency. According to Jon Sigler, TEU in ServicNow, these capabilities allow organizations to predict and improve the performance of agents prior to deployment, adding an essential layer of foresight to the development ​cycle of ​actors.

“For IA officers to have real commercial value, organizations ​need maximum confidence ​in their performance and ROI.”

As ServicNow said, this association ensures that companies can deploy IA agents with ​confidence, knowing that they can adapt to the changing needs of businesses.

What about the Azure AI roadmap?

Meanwhile, Microsoft has made more progress to reveal in Microsoft Build 2025 in Seattle. These include multi-agent orchestration updates through the ​semantic kernel agent framework, already available in general. Organizations such as KPMG use the semantic core to ​reduce the complexity of development by coordinating ​workflows between specialized actors.

In addition, Microsoft ​has published a report entitled DIY GenAI: Customization of A Generative ​for Unique Value, which ​provides information from over ​300 technology leaders. According to Microsoft, the results ​point out that customized models, ​either through ​an increased ​generation ​of fine adjustment or ​recovery (RAG), are at ​the ​heart ​of business differentiation in the AI era.

In short, Azure AI Foundry not only follows the fast evolution of AI, but actively forms. With intelligent orchestration, performance ​optimization and the safety of the agents on the platform, companies ​finally have a coherent system to deploy AI that feels like a team partner, not just a ​tool.

As AI continues to evolve from ​a useful assistant to a full ​partner, Azure ​AI Foundry is the platform of choice for those who are ready to lead the future of smart automation.