MicrosoftOf multi -agent framework, autogen, acts as a backbone for many enterprise projects, especially with its release Autojan v0.4 in January.
However, the company aims to harmonize all its agent framework offers and to expose further observation capabilities. Microsoft has issued an agent framework on a public preview, which will now be the company’s only archery and agent framework.
Microsoft said that Autojan and Cementic Dana will “remain in the rehabilitation mode, which means that they will not be new to the new feature but they will continue to receive Big fixes, security patches and stability.”
“However, for future work work, the roadmap is focused on the Microsoft Agent framework, and consumers should plan to migrate to the benefits of open quality, stability and the Ezore AI Foundry integration,” the company said in the venture bat.
The company assured that the current work burden on autojin or cementary kernel would be safe as “no breaking changes have been planned.”
The move to stabilize Microsoft’s agent framework shows the company’s strategy agent in the AI. The company’s purpose is to observe the framework of building agents and to close the data protection, enabling the company to create agents in one place, after deployment.
Agent Framework and Foundry
The agent framework stabilizes the workload of AI in a single SDK, which combines the capabilities of both cementary kernels and autogen, which allows users to build AI agents, manage multi -agent deployment and set up system.
Many developers and businesses are rapidly experiencing AI agents experience and adoption, but a lot of capabilities are needed to combine a lot of capabilities.
“The really interesting thing about what we are releasing this week is that there are many capabilities to help people more successfully build and manage agents, which, of course, allows them to be powerful.” “But, one that also ensures that they are reliable by giving you tools, you know, observe their behavior and new guards to help them stay at work.”
New framework offers five abilities for businesses for building AI agents:
Local experience before deployment to Azure AI Foundry
API integration through Open API and Run Times with A2A and MCP connection
Magic Use another archetypes agents
Reduce converting context in platforms
Create a multi -agent system in various agent platforms such as AI Foundry, M365 Cooplot or others
Microsoft is also included in its cloud -based foundry agent service, such as agent framework services, such as multi -agent workflows.
Safety, security and monitor
Bird said a difference for the agent framework is in the AI features responsible. Microsoft added:
Pressing the task, which keeps agents associated with tasks
PII detection, which warns administrators if any agent accesss sensitive data
Instant shields that help to prevent the injection immediately and highlight the behavior of a dangerous agent.
Bird said, “For the businesses with agents need to think about, I believe there are three important types.” “A number is the standard of a one agent, what does it actually work and what it is asking and completing it. Secondly, there is security, new types of risks such as traditional security and immediate injection attacks or sensitive data – and the third thing is the management of the third thing, because there are thousands of things to have access to work.
Microsoft will contribute to open lamery standards for observation. Through the AI Foundry, the developer can track quality, performance and price with agents created on the agent framework. The AI Foundry offers Open Limmetry Observing Agents Created with another framework and not only the agent framework.
An agent framework in all
Autojan competed with other agent builders and Multi -agent framework By LangchenFor, for, for,. Crew And llamainDex And there is no doubt that the agent’s framework will also be.
Microsoft is not the only one that is hoping to bring all required tools to build, deploy and monitor AI agents. Moving towards Langchin Offer these tools Even its purpose is to release 1.0. Since agents become more everywhere in enterprises, more platforms may consider access to building, deployment and observation tools in one.