Agentic Migrations and Modernization: How the Azure Migrate Agent Retains Your Intent Alive Finish to Finish

Whats up Of us!

You probably have ever tried to maneuver a couple of hundred VMs, a pile of databases, and a few net apps from on-prem to Azure, you already know the arduous half will not be the tooling.

The arduous half is maintaining context, intent, and momentum alive throughout weeks of planning, hand-offs, and choices. In session MAIS15 on the Microsoft Azure Infra Summit 2026, Ankur Gupta (Senior Product Supervisor on the Azure Migrate workforce) walked us by way of the brand new Azure Migrate agent, an AI layer that sits on prime of Azure Migrate and carries your intent from “I’ve an concept” all the best way to “the touchdown zone is deployed.”

Ankur opened with a line that caught with me. Infrastructure complexity has far outpaced human scale. Now we have MySQL right here, PostgreSQL there, net apps, storage gadgets, networking gear, a number of dashboards, a number of alerts, and we’re all anticipated to maneuver quicker than ever, with fewer errors. In brief, migrations hardly ever fail as a result of somebody picked the flawed instrument. They fail as a result of the system between the levels breaks.

Right here is why the agentic strategy issues for the parents within the trenches:

• It retains context throughout the whole lifecycle, so the intent you set on day one remains to be the intent at execution.
• It guides you when you find yourself caught, as an alternative of leaving you to determine which of three discovery strategies is the appropriate one.
• It compresses duties that used to take days of study (assume side-by-side enterprise circumstances) into a couple of hours.
• It connects IT ops, architects, and builders by way of a single thread of knowledge, together with a clear handoff to GitHub Copilot for code work.
• It builds on the Azure Migrate portal you already know, so nothing you’ve got discovered goes to waste.

That final level is essential. The portal doesn’t go away. The agent is a layer on prime. You possibly can nonetheless do every part you do immediately.

Azure Migrate has all the time been Microsoft’s hub for uncover, assess, and migrate. What Ankur confirmed at MAIS15 is the subsequent evolution. Azure Migrate is turning into a migration management aircraft that spans the entire lifecycle (Resolve, Plan, Execute), and the Azure Migrate agent is the conversational, guidance-oriented layer that ties all of it collectively.

In Ankur’s phrases, the agent is academic and guidance-oriented. You ask one natural-language query, like “how ought to I plan transferring my VMware workloads to Azure,” and also you get the subsequent steps you truly must take. Behind the scenes, the agent is doing three issues very properly.

1. It maintains state throughout the whole lifecycle. Preferences you set early stick with you.
2. It carries context throughout discovery, plan, and execute. You possibly can leap round, repeat steps, change your thoughts, and the agent remembers what your aim was.
3. It recommends the appropriate subsequent transfer based mostly on what it has discovered about your intent.

That is the guts of the “agentic” half. The agent will not be a chatbot grafted onto a portal. It’s a stateful workflow runner that remembers you.

The session walked by way of a full VMware-to-Azure situation, and the movement is price seeing as a result of it exhibits how the items snap collectively.

The agent helps three discovery strategies immediately: appliance-based discovery, RV Instruments, and the brand new Azure Migrate collector. The collector is the brand new light-weight choice. It ships as a set of PowerShell scripts you run on a machine that may attain your vCenter, it produces a zipper file, and also you add that zip to your Azure Migrate undertaking. No equipment to deploy, no inbound community plumbing.

As soon as the stock lands, the agent reads it. Ankur requested for a abstract and obtained a card exhibiting 207 VMs, 177 SQL databases, one PostgreSQL occasion, and a few net apps. He then requested for all servers with an out-of-support OS, obtained an inventory of round 50, and tagged them proper contained in the dialog so he may refer again to them later.

Subsequent got here the enterprise case. Ankur requested the agent to generate one based mostly on a modernize choice. A couple of minutes later, he had Azure price, on-prem price, and projected financial savings. Then he requested for a second enterprise case for raise and shift, and a side-by-side comparability. The agent ran it, confirmed that the on-prem price within the lift-and-shift comparability was greater and that lift-and-shift TCO financial savings had been truly greater in that particular situation, and gave him the info factors he wanted to deliver the choice to management.

From there, Ankur moved to utility evaluation, this time again within the portal. He created an evaluation for 2 apps (Airsonic and Elements Limitless), let the high-confidence plan run, and obtained a modernize advice with 100% readiness and a goal price of about $580 per 30 days, plus an emissions estimate of 32 kgs of CO2. App Service for the online tier, Azure Database for PostgreSQL for the info tier. Each had been flagged “prepared with situations,” with clickable hyperlinks into why.

Then got here one in all my favourite components. Ankur linked GitHub for a Copilot Evaluation, which provides code-level insights on prime of the infrastructure readiness evaluation. The system recalculated, and the migration effort estimate sharpened up.

Lastly, the agent constructed a wave plan from the evaluation, then generated a platform touchdown zone aligned with Azure greatest practices. He may ask the agent about chosen defaults, request adjustments to the deployment mechanism, swap in a third-party firewall, or apply naming conventions. The agent produced a downloadable Infrastructure-as-Code template and handed it to a cloud architect, who refined it of their IDE utilizing GitHub Copilot.

That final handoff is the bridge between Azure Migrate’s planning world and the developer world.

So the place does this truly repay? Just a few eventualities stood out.

• Pitching the enterprise case to management. Ankur framed the demo round “I must pitch a migration proposal to the planning committee.” Producing modernize, lift-and-shift, and Azure VMware Resolution enterprise circumstances was once days of spreadsheet work. With the agent, it’s hours.
• Cleansing up legacy debt. Tagging out-of-support servers in a single conversational step allows you to plan upgrades with out exporting CSVs and slicing them by hand.
• Blended estates with net apps and databases. The agent surfaces App Service and Azure Database for PostgreSQL targets, offers SKU suggestions, and flags the warnings price investigating.
• Closing the IT-to-developer hole. The GitHub Copilot Evaluation and the IaC handoff to the IDE means builders and designers work from the identical context.
• Decreasing intent drift on lengthy migrations. Multi-week journeys lose their unique intent. The agent remembers.

In brief, the ROI right here is measured in calendar time, not simply {dollars}. And truthfully, in fewer late-night calls when one thing goes sideways as a result of no one remembered the unique determination.

Tradeoffs price flagging: the agentic capabilities are touchdown in preview, and outputs are advisory. You continue to want human overview, testing, and governance on each advice. That’s by design.

Here’s a sensible onramp.

1. Rise up an Azure Migrate undertaking within the Azure portal if you don’t have already got one.
2. Decide a discovery methodology that matches your surroundings. Should you can’t deploy an equipment, attempt the brand new collector. Obtain the PowerShell scripts, run them from a number that may attain vCenter, and add the zip.
3. Deliver within the Azure Migrate agent from contained in the portal and ask it to summarize your found stock.
4. Generate a minimum of two enterprise circumstances (modernize and lift-and-shift). Evaluate them.
5. Run an utility evaluation on a small, consultant set of apps. Join GitHub and add a Copilot Evaluation for code-level perception.
6. Ask the agent to construct a wave plan and a platform touchdown zone template, then push the IaC to your repo for the architects.

Begin small, construct the muscle, and scale out.

Preserve Studying on the Summit

Catch the total Microsoft Azure Infra Summit 2026 session playlist right here Microsoft Azure Infra Summit 2026

Cheers!

Pierre Roman