Prompt Flow's Retirement Notice Is a 12-Month Migration Warning, and the Real Work Starts Now, Not in April 2027
On April 30, Microsoft formally told the world that Prompt Flow is going away. The retirement announcement gave teams twelve months — until April 20, 2027 — before Prompt Flow enters read-only mode and existing flows keep running but nothing new gets built. Feature development stopped April 20, 2026. Only security patches and critical bug fixes between now and retirement. That is the announcement in plain language, and it is accurate, and if you stop reading here you will have the gist of what Microsoft published. You will also have completely missed the point.
The point is not that a product is being retired. Product retirements happen. The point is that the AI workflow tooling consolidation inside Microsoft is now concrete enough to have a real migration problem attached to it. Prompt Flow was never just a flow-authoring visualizer. For enterprise teams building RAG pipelines, evaluation frameworks, multi-step agent loops, and prompt experimentation workflows inside Azure, it was the thing that made Foundry and Azure Machine Learning feel like a platform instead of a model API with extra steps. Its retirement is the clearest signal yet that Microsoft Agent Framework 1.0 — which reached general availability on April 3, 2026 — is the designated destination, and the twelve-month window is not a suggestion.
The migration path Microsoft has published is a rebuild, not a port. The announcement says it explicitly: "Migration may not always be like-for-like." Visual authoring experience migrations may require redesign. Custom tool registrations may need reimplementation. Evaluation pipelines built inside Prompt Flow may have no direct counterpart in MAF and may need to be rebuilt from scratch. Teams that treat this as a find-and-replace exercise are going to have a very uncomfortable Q2 2027.
The practical sequence Microsoft recommends is correct and worth following precisely: audit every Prompt Flow usage across the organization first, then migrate production-critical flows first, validate side-by-side, complete rollout before the retirement date. That sounds obvious. It is also the sequence most teams will skip because auditing is boring and migration is exciting. The teams that do the audit properly in the next thirty days will know their actual scope. The teams that wait will discover scope the hard way, under time pressure, with stakeholders watching.
The audit question is more specific than "how many flows do we have?" It is: which flows are in production? Which are using VS Code extensions versus the web authoring experience? Which are using evaluation capabilities that MAF will need to replicate? Which are calling custom tools that exist only inside a Prompt Flow context? Which are running on a schedule or triggered by external events that have operational dependencies? The answers define the migration scope, and scope defines whether you need three months or nine.
One underappreciated timing detail: the retirement workbook may not show Prompt Flow for up to two weeks after the announcement. Organizations with automated retirement tracking may have a silent gap before the deprecation registers in their dashboards. This is not theoretical. Microsoft acknowledged it in the announcement. Combined with the fact that feature development already ended April 20 — before the formal retirement announcement — teams should not assume their monitoring is current. Proactive manual audit is the only safe move. If you are a Prompt Flow shop and you have not run a manual audit this week, this paragraph is your call to action.
The Agent Framework 1.0 destination is worth understanding as a product, not just a migration target. MAF reached GA on April 3, 2026, in both Python and .NET flavors. It is a first-party agent runtime with hosted agents, typed contracts, OpenTelemetry tracing to Application Insights, and explicit support for multi-agent orchestration. It is not Prompt Flow with a new name. It is a different architectural model — more pro-code, more opinionated about how agents should be structured, and more aligned with the kind of production-grade governance that enterprise security and compliance teams actually require. The tradeoff is that MAF's power comes with more upfront design work. Prompt Flow's visual authoring was forgiving of loose thinking. MAF is less forgiving, which means the migration will surface architectural weaknesses that teams have been papering over.
The migration guide at aka.ms/PFtoMAFMigration and the code samples at aka.ms/PFMigrationCodeSamples are the starting points. They will not answer every question, because the honest answer to "how do I migrate my specific flow?" is "it depends on what your flow actually does." But they are the canonical references, and teams should treat them as the ground truth rather than interpolating from forum posts or third-party blog recaps of varying accuracy.
Here is the honest practitioner summary: this is a twelve-month problem that you need to start treating as a six-month problem, because the last three months of any migration timeline are always worse than the planning suggested. If you are running Prompt Flow in production, your job this week is the audit. Not the migration plan. Not the vendor evaluation. The audit. How many flows, which ones, which are critical, which have custom tooling, which have evaluation logic, which have operational dependencies. That document defines everything that follows. Everything else can wait. The audit cannot.
Sources: Microsoft TechCommunity, Learn — Prompt Flow, Microsoft Agent Framework 1.0