Agentic Migration: Where AI really helps – and where experience makes the difference

800 applications. No one in-house can say for certain which of these should be migrated, which modernised and which simply switched off. Sounds like an exceptional case? It isn’t.

The cloud has long since become part of decision-makers’ mindset and is firmly embedded in their way of thinking. What’s missing is rarely the will – but rather the big picture. Current market figures also show that the focus is increasingly on actual implementation rather than just strategy: according to the Flexera 2025 State of the Cloud Report, the metric ‘number of migrated workloads’ – used as a measure of success for cloud targets – has jumped from 36 per cent to 78 per cent year-on-year. Companies are therefore increasingly being judged on whether they are actually migrating – not just on whether a strategy exists. — Companies are therefore increasingly being judged on whether migration is actually taking place, not just on whether a strategy exists. We have been supporting such projects for years with our App2Cloud method, and in recent months we have started to deploy AI agents specifically in certain phases of this process. It’s time for an honest look at where this really makes a difference – and where experience and effective change management still make all the difference.

The App2Cloud framework at a glance

App2Cloud is not a single workshop, but a process that guides three workstreams in parallel through several phases:

• Strategy – from the vision to the concept & roadmap, right through to the operating model for cloud enabling and operational excellence.
• Environment – from discovery of the existing environment through foundation and landing zone design to ongoing cloud management.
• Applications – from assessment of individual applications through migration and modernisation to app optimisation during live operation.

Two themes cut across all three workstreams, and in our experience, they determine success or failure: project management and change management. One ensures that the project runs on schedule. The other ensures that, in the end, there is someone who is both willing and able to operate the result. Those who plan only the former end up with a technically flawless project that fails in production.

The challenge: taking stock as a full-time job

Before a single application can even be migrated, there is an unglamorous but crucial question to be answered: what do we actually have? In practice, this means weeks or months of interviews with business departments, maintaining spreadsheets, and painstaking detective work to uncover dependencies that nobody has documented anymore. This experience is borne out by current market figures: for 59 per cent of European companies, understanding application dependencies is the biggest hurdle, According to the Flexera 2025 State of the Cloud Report, understanding application dependencies is the biggest hurdle to migration — even ahead of assessing technical feasibility (52 per cent) and comparing the costs between on-premises and cloud solutions (41 per cent). It is precisely this lack of transparency that the automated discovery phase addresses. No wonder that many migration projects stall at this stage – or that, under time pressure, teams resort to a simple ‘lift-and-shift’ approach. The problem with this is that it merely shifts legacy technical issues to the cloud rather than resolving them.

Where the agents actually come into play

It is important to note that Agentic Migration does not replace this framework; rather, it specifically accelerates two of its phases.

In the discovery/assessment phase, agents take on the legwork that previously took weeks – automatically scanning the environment via cloud APIs, normalising the raw data, and providing an initial, reasoned 7R assessment (retire, retain, rehost, replatform, refactor, repurchase, replace) for each application, including a confidence score. This is pure speed, not magic: what is achieved in our traditional approach through technical discussions and expert input is delivered by the agent as an initial, machine-generated draft – although it is still reviewed and approved by a human.

During the Migration & Modernisation phase, the agents assist in deriving the wave plan, target architecture and the initial Infrastructure-as-Code frameworks – again, these are generated automatically but are continuously checked against guardrails and compliance requirements.

What is deliberately not handled by agents: strategy, operating model, project management and – this is the key point we’ve learnt most clearly from real-world projects – change management. These are not data problems that can be solved through better automation. These are leadership and organisational issues.

What real projects have taught us

In our experience, when cloud migration projects fail, it is rarely down to the technology. An application that has been migrated technically correctly – that the team has migrated technically correctly – but is to be taken over by an operations team that has never worked with cloud-native infrastructure before, still poses a risk. A 7R recommendation that is technically sound but has not been explained to anyone in the business unit will be boycotted or delayed by that unit during implementation – regardless of whether it was drawn up by a human or an agent. And this is precisely where the real leverage for agent-driven acceleration lies: if the baseline assessment takes just a few weeks rather than three months, there is simply more time left in the project plan for operational preparation and change management – rather than teams having to rush through both at the end under time pressure. More time for the human aspects, without compromising on care – that is the difference that AI agents really make here.

Let’s take a fictional but typical example: a medium-sized industrial company with around 300 applications running on AWS, Azure and its own server room. The environment is scanned within hours; a stable billing service is identified as a prime candidate for rapid rehosting; a CRM system deeply integrated with legacy systems is flagged for refactoring; and the agent recommends decommissioning a rarely used reporting application. Because this assessment no longer takes months, there is sufficient lead time to work out the future operating model with the teams concerned – rather than presenting it to them just before go-live.

Agentic Migration does not, therefore, replace App2Cloud; rather, it specifically accelerates the phases that used to take the most time. Anyone who believes that AI agents would render change management and operational preparation superfluous has either never seen a migration project through to completion, or does not remember the last one properly.