Microsoft Agent Framework 1.0 Migration Is Harder Than Billed

What happened

Microsoft Agent Framework reached general availability on 2 April 2026, and Microsoft is positioning it as the home for its agent tooling going forward. Semantic Kernel is not being switched off yet. Microsoft has committed to fixing critical bugs and security issues in Semantic Kernel v1.x for at least one year after Agent Framework's GA date 1. Most new capabilities, including the agent harness and CodeAct, are being built for Agent Framework rather than for Semantic Kernel 1.

The official message is that the older framework still works while the newer one gets the new features. Developer reports from about six months after launch describe the move between them as a significant rewrite, not a drop-in upgrade 2.

The support window

The one-year commitment is a minimum, not a scheduled end date, and it covers critical bugs and security fixes only 1. Teams running Semantic Kernel in production can therefore expect patches for serious problems at least into early April 2027. They should not expect the platform to keep improving.

That creates a gradual kind of pressure. Code built on Semantic Kernel keeps running, but it falls further behind with each feature Microsoft ships only in Agent Framework. For organizations that adopted Semantic Kernel to stay close to Microsoft's direction, this distinction matters more than the formal support terms.

Where the migration breaks

The most basic change is that Semantic Kernel's central Kernel object goes away completely in Agent Framework 2. Questions about what takes its place are among the first ones developers raise when planning a move 1. Because the Kernel organized services, plugins and invocation in many Semantic Kernel codebases, losing it changes the overall structure of an application, not just a few lines.

Teams coming from AutoGen have their own problems. Patterns built on AssistantAgent and GroupChat have to be rebuilt as graph-based Workflows 2. That is a different way of modeling how agents coordinate, so the port is closer to a redesign.

Developers also report breaking changes in several other areas [2]:

• Session state: how conversation state is stored has changed.
• Telemetry tags: existing dashboards and alerts may stop matching.
• Tool schemas: tool definitions need updating.

According to those accounts, the cumulative effect has been dedicated migration sprints rather than incremental updates 2.

Gaps the announcement didn't mention

.NET developers have pointed to missing or weaker features. They note there is no text-chunking support, which matters for retrieval-style workloads, and that tool approval is less capable than what Semantic Kernel offered 2. One engineer described the framework as "designs implemented poorly." He also claimed Microsoft itself had not adopted it internally, and chose to build on Microsoft.Extensions.AI instead 2.

That last claim comes from one developer's comments and should be treated as an opinion, not an established fact. Still, the decision is worth noting. Microsoft.Extensions.AI is a lower-level abstraction, and some teams appear to prefer building their own orchestration on top of it rather than adopting a full framework that is still settling.

How the two accounts fit together

The two accounts mostly agree on the facts but differ in tone. The migration-guide framing treats the change as a planned transition with a clear support runway 1. The developer reports focus on what that transition costs in practice 2. Neither disputes that Agent Framework is where Microsoft's investment is going. The disagreement is about how ready it is to replace what came before.

Analysis: plan for a project, not a version bump

The evidence points to a familiar pattern. A vendor's "1.0, production-ready" label describes the API's stability and Microsoft's commitment. It does not say how easy it is to move existing systems onto it. Agent Framework may well become the right foundation. For now, migrating to it is a real engineering project.

A practical approach for teams on Semantic Kernel or AutoGen:

1. Treat the one-year window as time to plan. Use it to inventory every use of the Kernel object, the AutoGen agent patterns, telemetry and tool definitions.
2. Check for missing pieces first. If you rely on text chunking or strict tool approval, confirm how you will replace them before committing to a date.
3. Consider staging the move. Building new work on Agent Framework while existing services stay on supported Semantic Kernel code may be easier than one large port.
4. Keep alternatives in view. For teams that mainly need model abstraction, building directly on Microsoft.Extensions.AI is an option at least one developer has taken.

Agent Framework is clearly Microsoft's future direction. How quickly a given team should adopt it depends on how much of its system needs to be rebuilt along the way.