What Azure Catalyst is
Microsoft calls Azure Catalyst a new, unified experience for agentic modernization across Azure and GitHub. It combines AI-powered discovery, planning, migration and modernization, so IT teams and developers can work across the whole lifecycle.
Microsoft's Ignite materials add some naming and status context:
- The Azure Catalyst agent was formerly the Azure Copilot migration agent. It is now generally available.
- New preview capabilities extend that agent from migration planning into deeper modernization and execution.
- Other coverage describes the service as a rebrand and expansion of Azure Migrate. Microsoft's database blog says Azure Catalyst (formerly Azure Migrate) now covers applications and databases together.
- Virtualization Review says it combines Azure Migrate and GitHub Copilot modernization capabilities with AI agents.
In short, existing Azure Migrate users should expect the tooling they know to move under a new name and a wider scope.
What's new in the workflow
Microsoft's announcement lists these new capabilities:
- Estate insights.
- AI-powered application discovery.
- Modernization strategy with cost comparisons.
- AI-generated migration waves.
Execution covers applications, containers, storage, networking and databases. It builds on existing Java and .NET modernization, plus migration support for VMware, Hyper-V and physical servers.
Virtualization Review adds that assessment now extends to containers, SQL databases, MongoDB, Oracle databases, file servers and network-attached storage. It also says organizations can use GitHub Copilot to assess and transform .NET and Java applications alongside their databases.
Two items are explicitly previews:
- Modernizing applications and databases together.
- Migrating from BizTalk to Azure Logic Apps.
Microsoft's blog gives no dates or eligibility rules for these previews. BizTalk shops should check the preview terms before building a plan around them.
Governance and security claims
Microsoft says governance rulebooks and bring-your-own-model support let customers set the standards and guardrails agents follow. Reusable runbooks and skills are meant to make successful patterns repeatable.
Catalyst also works with Microsoft Security to rank risk by severity and exposure. It flags vulnerable code and databases, and GitHub Copilot can help fix them during modernization. Virtualization Review describes this as security analysis built into modernization planning, with developer tools for addressing problems during application transformation.
This is a product description, not a guarantee. The announcement doesn't explain how agent-generated changes are reviewed or approved. Teams should confirm those controls and still test, review and approve changes through their normal change process.
The numbers, and how to read them
Microsoft's reported figures:
- 1.4 million applications modernized.
- Levi Strauss has deployed more than 1,000 agents across its enterprise systems.
- Microsoft Quantum and Xbox achieved 80% faster app modernization.
- Intune upgraded a key library 75% faster than anticipated.
- GitHub ported the GitHub Copilot agentic harness, 800,000 lines of production code, from TypeScript to Rust.
The post gives no baselines, methodology or independent validation for any of these. "Faster than anticipated" depends entirely on the original estimate. They show what is possible inside Microsoft, which calls itself "customer zero", but they aren't a forecast for your estate. The claim that months-long work now takes weeks or days should be treated the same way until you've run a pilot.
AI infrastructure: isolation, networking, context
The other half of the announcement targets running agents on Azure Kubernetes Service (AKS) and Microsoft Foundry.
| Feature | What Microsoft says it does |
|---|---|
| Agent Sandbox extension for AKS | Runs agent-generated code in managed sandboxes with VM-level isolation per session, pre-warmed pools for fast startup, and a serverless option |
| Agent Grid | Connects agents to other agents, models, tools, data and Azure services over authenticated, encrypted links, with identity-driven policies and default-deny network isolation |
| Context caching | Long-running agents reuse processed prompt context instead of recomputing it on the GPU, with the cache kept durable and controlled in your tenant |
| AKS AI Manager | Matches each model's needs to GPU and VM performance, topology and real-time fleet availability, then places the deployment accordingly |
| Foundry-to-AKS path | Open-weight models and agents move from Foundry to AKS clusters, from cloud to edge, under consistent policy and governance |
The post doesn't give release status, supported configurations or benchmarks for these features. It says they make agents faster, cheaper and more secure, but offers no comparative latency, cost or security test results. The open-source layer is named as Kubernetes, vLLM and Ray.
Default-deny networking and per-session VM isolation are sensible design choices for code that an AI writes and then runs. It's a bad idea to let such code run with ambient access. Whether the implementation lives up to this needs testing, not a blog post.
Distributed AI and Windows Server 2028
Microsoft says Azure Local and flex nodes for AKS extend Foundry and Kubernetes compute into distributed environments, with Azure Arc handling governance. Windows Server 2028 is named as now in public preview in the blog's summary, positioned for running AI inference alongside business-critical applications and data. The post gives no hardware requirements or deployment steps for it. Windows PCs are described as bringing AI closer to users through local processing.
The flex nodes caveat
Flex nodes are the clearest preview boundary. Microsoft Learn describes them as extending AKS to user-managed virtual machines and bare-metal hosts, so you can add existing capacity without moving it into a standard node pool. Typical uses include capacity outside the cluster's Azure region, bare-metal hardware, lab and test, and data-residency needs.
Microsoft's documentation says these points about the preview:
- Previews are provided "as is" and "as available", are excluded from SLAs, and aren't meant for production use.
- Flex nodes run on Linux hosts you provide. You operate the host, its operating system, identity, network connectivity and workloads.
- Authentication options are Azure Arc managed identity, Azure VM managed identity, or a service principal.
- A cluster containing a flex node pool can't be stopped.
- Flex node pools can't be created or managed through the Azure portal.
- The documented deployment assumes Unbounded-Net pod networking and existing Layer 3 connectivity between the AKS-managed and flex node networks.
The Ignite blog ties flex nodes to "greater flexibility from development through production". The documentation says production use isn't recommended while the feature is in preview. Treat flex nodes as something to evaluate, not as ordinary AKS capacity.
Customer examples
Microsoft cites Rockwell Automation, which uses AI in factories to help engineers diagnose equipment issues and investigate product failures. A Rockwell executive says the aim is to adapt what works in one plant for the next. It also describes COSMIC, a shared AKS platform from Microsoft 365 engineering that handles infrastructure centrally so developers avoid Kubernetes operations. Both are Microsoft-supplied examples, not independent evaluations.
What to do with this
- Separate GA from preview. Per Microsoft's Ignite material, the Catalyst agent is generally available. The new execution, app-plus-database and BizTalk capabilities are previews, and flex nodes are a preview too.
- Run a pilot on a low-risk app. Pick a .NET or Java workload, compare the agent's plan and cost comparison against your own estimates, and record the actual effort.
- Define your guardrails first. Governance rulebooks and bring-your-own-model are only useful if you've written down your standards.
- Keep human review in the loop. Run agent-generated code changes and vulnerability fixes through your usual testing and approval gates.
- Plan for the shared-responsibility split before trying flex nodes: you own the host OS, patching, identity and network path.
Bottom line
Catalyst packages migration, modernization and security triage into one agent-assisted workflow spanning IT and developers. That's a real consolidation, and the Azure Migrate lineage gives it substance. The efficiency claims are unproven outside Microsoft, and several of the pieces that matter most are still in preview. Test it before betting a modernization roadmap on it.
References
- Introducing Azure Catalyst agentic modernization and new AI infrastructure innovations - Microsoft Azure Microsoft Azure · 2026-11-17T18:35:00+00:00
- What are flex nodes for AKS? (preview) - Azure Kubernetes Service | Microsoft Learn learn.microsoft.com
- Support policy for flex nodes in AKS (preview) - Azure Kubernetes Service | Microsoft Learn learn.microsoft.com