Semantic Kernel will be deprecated?

[gsubiran]

Today I had a meeting with a Microsoft partner regarding some GenAI projects. I mentioned our tech stack, which includes Semantic Kernel with C#, among other things.

To my surprise, the partner said that Semantic Kernel is a project that Microsoft plans to discontinue. This really surprised me, as I see it as a growing project with a lot of activity.

That's why I'm writing this—to find out how true this is and whether we should consider switching to other orchestrators or if it's still worth investing in this wonderful project.

[rajyraman]

@gsubiran - See https://devblogs.microsoft.com/semantic-kernel/semantic-kernel-and-microsoft-extensions-ai-better-together-part-1/

[moonbox3]

@rajyraman the dev blog link you posted doesn't answer the question -- a lot of SK abstractions have moved to the more central location of being included in the Microsoft.Extensions.AI package. That doesn't mean SK is getting discontinued.

@gsubiran could you provide some more context about your partner saying "that Semantic Kernel is a project that Microsoft plans to discontinue?" SK isn't getting discontinued.

[gsubiran]

@rajyraman the dev blog link you posted doesn't answer the question -- a lot of SK abstractions have moved to the more central location of being included in the Microsoft.Extensions.AI package. That doesn't mean SK is getting discontinued.

@gsubiran could you provide some more context about your partner saying "that Semantic Kernel is a project that Microsoft plans to discontinue?" SK isn't getting discontinued.

@moonbox3 I agree with you—relying on centralized abstractions doesn’t necessarily mean that SK is being discontinued. Just to give you more context on the conversation where the partner mentioned this: in my organization, we’ve built several POCs using Semantic Kernel and all the tools it has offered since its early days. Recently, we had a meeting with a Microsoft partner to discuss a typical GenAI project we want to move forward with—essentially, a Teams chatbot that answers queries based on a knowledge base.

During the meeting, I shared our tech stack to check whether we were aligned with the technologies commonly used within the Microsoft ecosystem. Everything seemed fine until I mentioned Semantic Kernel. At that point, they said they agreed with all our choices except that one. According to “conversations they had with Microsoft,” Semantic Kernel was going to start being phased out.

This caught me by surprise, especially given how active the project still is and how many new features are reaching GA status. Unfortunately, they didn’t provide much more context beyond that.

If you happen to have any insights into this, I’d really appreciate some clarification. Is there any official roadmap or communication from the team regarding the long-term plans for Semantic Kernel? Also, if there are any other recommended technologies within the .NET ecosystem that Microsoft is prioritizing for GenAI workloads going forward, any advice would be very welcome.

[markwallace-microsoft]

@gsubiran There are no plans to discontinue Semantic Kernel. Instead, we are actively collaborating with other teams at Microsoft to enhance the AI developer ecosystem. Here are some key initiatives:

1. Partnership with the .NET Team on Vector Data
   We have partnered with the .NET team to develop the Microsoft.Extensions.VectorData.Abstractions. This work was carried out in the Semantic Kernel repository, alongside the development of most of the Vector Store implementations.
2. Integration with Microsoft.Extensions.AI.Abstractions
   We’re also collaborating with the .NET team on Microsoft.Extensions.AI.Abstractions, which is now fully supported in Semantic Kernel. Our team is actively contributing to this project. For new AI primitives, such as real-time support, the plan is to introduce them via M.E.AI. One key advantage of this partnership is that the MCP integration required no changes to Semantic Kernel’s codebase, as both our packages and the MCP .NET package are built on M.E.AI.
3. Convergence on a Unified Agent Framework
   We’ve joined forces with the AutoGen team to create a single, unified Agent Framework for Microsoft. More announcements on this initiative will be shared in the coming weeks. As always, our goal is to safeguard your investments in Semantic Kernel.

The move toward using Microsoft.Extensions reflects feedback from the developer community. Having more focused abstractions as official Microsoft extensions makes it easier to drive adoption and improve interoperability.

It’s important to note that these collaborations do not signal a reduction in our investment in Semantic Kernel. Our team remains fully committed to supporting and growing our developer community.

[gsubiran]

@markwallace-microsoft Thanks a lot for the detailed response and for sharing this information.

Honestly, I also found the partner’s comment quite surprising—that’s why I came here to see if there was something I might be missing. But now it's clear that this is not the case.

I'm not sure what led the partner to make that statement, but in any case, I really appreciate the clarification and all the work your team is doing to strengthen the ecosystem around Semantic Kernel.