For years, the bottleneck in large cloud migrations was not moving servers. Replication tools solved that long ago. The real constraint was everything around the move: discovering what exists, figuring out what depends on what, building a business case, designing landing zones and networks, grouping applications into waves and coordinating dozens of teams. Those tasks consumed months of expert time and spreadsheets. Since late 2025, the major cloud providers have released agentic AI tools that automate large parts of this work. Used well, they can compress planning from months to weeks. Used carelessly, they can produce confident plans built on incomplete data. This article explains what the new tools actually do, how they change the migration factory and how to adopt them with the right controls.
What Changed: From Migration Tools to Migration Agents
Traditional migration tooling was a collection of point products: a discovery collector, an assessment calculator, a replication engine and perhaps a dashboard. Humans stitched them together. The new generation uses AI agents to orchestrate the whole workflow and to reason over messy, unstructured inputs.
Three releases illustrate the shift:
- AWS Transform reached general availability for VMware workloads in May 2025 and, in December 2025, added agentic capabilities that discover and prioritize workloads, ingest third-party inventories and unstructured documents, map dependencies, generate wave plans, design hub-and-spoke or isolated networks, translate configurations from NSX, Palo Alto, FortiGate and Cisco ACI environments, and migrate servers from VMware, Hyper-V, Nutanix, KVM and bare metal. In a June 2026 post, AWS described it as a unified workflow spanning assessment, wave planning, landing zone setup, network conversion, containerization and rehosting, accessible from coding assistants through an MCP server.
- Azure Copilot migration agent entered public preview in March 2026. Microsoft positions it as bringing AI into discovery, assessment, planning and deployment, giving early visibility into inventory, cost and modernization candidates.
- GitHub Copilot modernization agent, also in public preview from March 2026, orchestrates code assessments across applications, creates modernization plans and performs framework and runtime upgrades, closing the gap between infrastructure migration and application change.
Google Cloud is moving in the same direction, positioning its migration programs and tooling around AI-assisted assessment and modernization. Across vendors, the pattern is consistent: agents propose, humans review and approve.
Where AI Adds the Most Value in a Migration
Discovery and Data Reconciliation
Real inventories are inconsistent. RVTools exports disagree with the CMDB, application owners describe systems differently from monitoring tools, and critical knowledge lives in documents and runbooks. Agents are good at reconciling these sources, flagging conflicts and extracting facts from unstructured material such as architecture documents or business rules.
Dependency Mapping and Wave Planning
Grouping hundreds or thousands of applications into move groups is a combinatorial problem that humans solve slowly. AWS says Transform can surface dependencies hidden in firewall rules and produce wave plans that leads can edit, re-sequence and sign off. The value is not only speed; it is the ability to rerun the plan in minutes when an application owner changes a constraint.
Business Cases and What-If Analysis
Agents can model alternative regions, utilization assumptions and service mappings quickly, producing comparable scenarios rather than a single static estimate. This is especially useful when finance asks “what if we keep these workloads on-premises?”
Landing Zones and Network Translation
Network translation has historically been one of the most error-prone parts of a data center exit. Tools that translate micro-segmentation and firewall policies into cloud constructs, generate infrastructure-as-code and flag hard-coded IP addresses remove a significant source of delay, provided network engineers review the output.
Containerization and Code Modernization
For workloads that should be modernized rather than rehosted, agents can generate Dockerfiles, deployment manifests and framework upgrades. AWS reports customer examples of containerizing an application in hours rather than days, and Microsoft’s modernization agent targets automated runtime upgrades. These are vendor-reported results, so validate them on your own code before building plans around them.
Where Humans Still Decide
The most important design principle in all of these tools is human-in-the-loop control, and it should be in your process too. AI is weak in precisely the areas where migrations carry the most risk:
- Business context: an agent cannot know that a quarter-end close makes the finance systems untouchable in March, or that a particular plant cannot tolerate a weekend outage.
- Undocumented behavior: batch jobs that run once a year, hard-coded integrations and manual workarounds rarely appear in telemetry.
- Risk acceptance: choosing which risks to take, and when, is a governance decision, not a calculation.
- Contractual and licensing nuance: software licensing rules and support contracts often constrain where and how workloads can run.
- Organizational readiness: whether a team can operate its application in the cloud the day after cutover.
AWS’s own summary of its approach is that the AI proposes and the team decides. That should be the operating rule for any agentic tool in your migration.
How the Migration Factory Changes
A classic migration factory organizes people into discovery, planning, migration and validation teams with handoffs between them. Agentic tools reshape these roles:
- Discovery analysts become data curators. Instead of compiling spreadsheets, they validate agent-reconciled inventories and resolve conflicts with application owners.
- Planners become plan reviewers. They set constraints and priorities, review agent-generated waves and own the sign-off.
- Network and platform engineers review generated designs and code rather than writing everything from scratch, with peer review and policy checks in the pipeline.
- Migration engineers focus on exceptions. Routine rehosts run in automated waves; humans handle complex databases, clustered systems and applications that fail validation.
- Application owners engage earlier, because fast replanning makes their input immediately actionable.
The result is not fewer people in every case, but a different mix: more time spent on judgment, validation and stakeholder alignment, and less on data assembly.
Choosing the Right Tooling Approach
Most organizations will use the tooling of their target cloud, but the choice is not always that simple. AWS’s own guidance on agentic options for VMware workloads contrasts a packaged service, AWS Transform, with a custom multi-agent approach built on Amazon Bedrock. The packaged service suits straightforward rehosts and teams that want results quickly; AWS notes that core assessment and transformation features are offered at no charge. The custom approach, with a supervisor agent coordinating specialist agents and MCP connections to your own tools, suits complex replatforming, non-standard sources and mature cloud teams, at the price of higher setup effort and ongoing operations. Many enterprises combine both: a packaged tool for the bulk of standard servers and custom agents or scripts for the unusual cases.
When comparing options, evaluate:
- Source coverage: hypervisors, operating systems, databases and physical servers supported, and how well the tool handles your oldest systems.
- Input flexibility: ability to ingest your existing discovery exports, CMDB data and unstructured documents without re-collecting everything.
- Transparency: whether the tool explains why it grouped applications or sized instances the way it did, so reviewers can challenge its reasoning.
- Editability and export: whether plans, designs and code can be edited, exported as infrastructure as code and version-controlled outside the tool.
- Security and data handling: where artifacts live, which permissions the agents need and what audit trail they leave.
- Multicloud and hybrid fit: how the tool behaves if part of the estate is going to another cloud or staying on-premises.
Adopting Agentic Migration Tools: A Practical Playbook
1. Start With Data Quality
Agents amplify input quality. Before running assessments, collect discovery data from multiple sources, including performance telemetry over a representative period, and agree on a single application inventory with owners.
2. Run a Bounded Pilot
Select a slice of the estate, such as one business unit or one data center hall, and run the agentic workflow end to end alongside your existing process. Compare the dependency maps, wave plans and network designs against what your experts would produce. Measure planning time, rework and defects found in review. Include at least one awkward application in the pilot, such as a system with undocumented integrations or a clustered database, because the edge cases reveal far more about a tool’s limits than the easy servers do.
3. Define Review Gates
Decide which outputs require human sign-off: wave composition, network designs, security group rules, landing zone templates and cutover runbooks at minimum. Put generated infrastructure code through the same pull-request reviews and policy checks as hand-written code.
4. Govern Data and Access
Migration data includes IP addresses, hostnames, firewall rules and sometimes credentials. Check where the tool stores artifacts (AWS, for example, allows customers to bring their own S3 bucket), which identities it uses to act in your environments, and how its actions are logged. Grant migration agents least privilege, scoped to the accounts and resources in each wave.
5. Integrate With Your Landing Zone Standards
Generated landing zones and networks should conform to your existing architecture and policy baseline, not create a parallel one. Feed your standards into the tool where possible, and run policy-as-code checks on every output.
6. Keep a Decision Log
Record each significant agent recommendation, the human decision taken and the reason, especially when the team overrides the tool. This log becomes valuable audit evidence for regulated industries, and it reveals patterns where the tool consistently needs correction, which you can feed back as constraints in later waves.
7. Measure and Improve
Track effort per server or application, wave success rate, rollback rate, post-migration incidents and time from discovery to cutover. Use these metrics to decide where to extend automation and where to keep humans firmly in control.
A Worked Example: Planning 2,000 Servers in Weeks
Consider a hypothetical healthcare services company exiting two data centers with about 2,000 servers, mostly on VMware, and a contract deadline eighteen months away. In the past, its consultants estimated three months of discovery and planning before the first wave.
The company collects RVTools exports, a month of performance data, CMDB extracts and firewall configurations, and feeds them into an agentic migration tool. Within days, it has a reconciled inventory with flagged conflicts and a first dependency map. The migration team spends the next two weeks validating the map with application owners; they discover several undocumented integrations with a laboratory system and a yearly regulatory batch job that the telemetry missed. They encode these as constraints, and the agent regenerates the wave plan in hours.
Network engineers review the generated landing zone and translated firewall rules, correcting a handful of overly broad rules before approval. The first pilot wave of fifty low-risk servers migrates in week six, rather than month four. Complex systems, including the clinical databases, follow dedicated plans with human-led design. The overall program still takes more than a year, but planning effort and replanning cycles shrink substantially, and the team can absorb changes without derailing the schedule.
Pitfalls to Avoid
- Trusting plans built on thin data: a beautifully formatted wave plan is still wrong if discovery missed key dependencies.
- Skipping network review: automatically translated rules can be overly permissive or break subtle traffic flows.
- Letting tools define architecture: generated landing zones must align with your enterprise standards.
- Over-reading vendor metrics: published speedups come from selected case studies; measure your own.
- Neglecting change management: application owners and operations teams still need training, runbooks and support after cutover.
A final caution concerns the post-migration phase. Agentic tools concentrate on getting workloads to the cloud, but value is realized afterward: rightsizing, adopting reserved capacity or savings plans, tightening security posture and retiring the old infrastructure. Plan an optimization sprint after each wave, and use the same AI-assisted analysis to identify idle resources and modernization candidates. Migration programs that stop at cutover often find their cloud bills higher than the business case assumed, which undermines confidence in the next phase.
Agentic tools make migration planning dramatically faster, but they do not make it automatically right. Let AI propose, keep experts deciding, and invest in the data and review gates that turn speed into safe outcomes.
Conclusion: Faster Planning, Same Accountability
AI-assisted migration is one of the most practical, immediately useful applications of agentic AI in enterprise IT. Tools from AWS, Microsoft and others can now reconcile inventories, map dependencies, generate wave plans, translate networks and even modernize code. The organizations that benefit most treat these agents as accelerators inside a disciplined migration factory: they feed them good data, define clear review gates, govern their access and measure results. The work of migration shifts from assembling information to making decisions, which is exactly where experienced people add the most value.
How Hanoi Technology Can Help
Hanoi Technology runs AI-assisted migration factories that combine agentic tooling with expert review, from discovery and wave planning to landing zones, rehosting and modernization.
Further Reading
- AWS Transform adds new agentic AI capabilities for enterprise VMware migrations – AWS
- How AWS is using agentic AI to reinvent infrastructure modernization – AWS Migration & Modernization Blog
- AWS agentic AI options for migrating VMware-based workloads – AWS Migration & Modernization Blog
- New Microsoft AI Agents to Help Modernize Enterprises – AI Business