What you need to know
Prompt flow in Microsoft Foundry and Azure Machine Learning will be retired on 20 April 2027.
As part of this change, Microsoft recommends moving to Microsoft Agent Framework, which provides a production-ready foundation for building agent and workflow-based AI applications. Microsoft Agent Framework 1.0 is generally available for both Python and .NET. During the transition period, customers should begin planning and executing their migration so that existing applications, workflows, and deployments are moved to supported alternatives before the retirement date.
What is changing
After 20 April 2027, prompt flow components will no longer be supported.
This includes:
- the web authoring experience in Microsoft Foundry and Azure Machine Learning
- the VS Code extensions for prompt flow
- the prompt flow base image used in deployments
During deprecation period till 20 April 2027, prompt flow will continue to receive security and critical bug fixes, but no further feature development is planned.
Recommended path forward
Microsoft recommends Microsoft Agent Framework (GA since 3 April 2026) as the primary path forward for customers using prompt flow. The framework supports both agents and workflows, and is designed for production use with stable APIs and long-term support.
You can learn more here:
What customers need to do
- Review your current prompt flow usage
Start by identifying where prompt flow is currently used across your organization, including:
- active applications
- existing workflows
- deployed endpoints
- development or evaluation workflows
- operational dependencies tied to Prompt Flow
This helps you understand the scope of migration and prioritize the most important workloads first.
- Plan your migration target
For most workflow and orchestration scenarios, the recommended destination is Microsoft Agent Framework. If you also rely on evaluation and tracing capabilities, review the latest supported options in Microsoft Foundry as part of your migration plan.
- Rebuild and validate
Migration should include both:
- rebuilding the application or workflow in the new framework
- validating that the new implementation behaves as expected before production cutover
A phased approach is usually best: migrate one business-critical scenario first, validate it, and then expand to additional workloads. Migration is a multi-phase process that includes audit, rebuild, validation, operational migration, and cutover.
- Update deployments and operations
If your solution uses prompt flow deployments today, plan to move those deployments to supported alternatives before retirement.
What to expect during migration
Migration may not always be like-for-like
Customers should plan for migration as an application update, not just a platform switch.
In some cases, especially where prompt flow was used heavily as a visual authoring experience, migration may require some redesign of the solution rather than a direct one-to-one replacement. Customers should not assume a simple visual replacement and may need to preserve the business outcome rather than recreate the exact previous experience.
Start with the highest-value scenarios
A practical approach is to:
- identify production-critical flows first
- migrate the most important scenarios before lower-priority ones
- validate outputs side-by-side where needed
- complete rollout well before the retirement date
Support and help
If you need additional guidance:
- Review the migration guide and code samples
- Learn more about Microsoft Agent Framework
- If you need technical assistance, use your standard Azure support channels:
- Get answers from community experts in Microsoft Q&A
- Create a support request.
- For Issue type, select Technical.
- For Subscription, select your subscription.
- For Service, select My services, then select Machine Learning or Microsoft Foundry.
- For Summary, type a description of your issue.
- For Problem type, select prompt flow.
- For Problem subtype, select your appropriate problem area.
Learn more about service retirements that may impact your resources in the Azure Retirement Workbook. Please note that retirements may not be visible in the workbook for up to two weeks after being announced.
Microsoft