Agentic AI: a new opportunity for industrial legacy
There is one question that comes up more and more frequently in conversations with industry executives: What do we do about those systems that have been running the plant for twenty or thirty years and which nobody dares to touch?
The answer is not straightforward, given that, in many cases, we are talking about platforms that continue to operate reliably, that control critical processes and that have built up years of knowledge about how the company actually operates.
For years, the modernisation of these legacy systems has occupied an awkward position on the agenda of large industrial companies. They all recognised its importance, but there was almost always a more pressing priority: maintaining production, rolling out a new ERP system, strengthening cybersecurity or meeting a regulatory requirement. The system was still working, and that is precisely why the decision has been put off.
Today, that logic is beginning to break down. Not because legacy systems have suddenly stopped working, but because the cost of maintaining them is increasingly being measured not so much in terms of technology as in terms of business opportunities that cannot be pursued. And because artificial intelligence – particularly as it evolves towards agent-based models – is radically altering the economic, temporal and risk equation associated with its modernisation. Furthermore, we face the problem that some of that knowledge is not held within the organisation; at best, it is contained within the code or in the minds of the few people who still know how to interpret it.
Agentic AI makes it possible to tackle the technological debt that has built up over decades, not only because it enables a more agile transition, but also because it presents an opportunity to reinvent the way the industry operates.
Legacy is not just about seniority: it is about restriction
A system does not become legacy simply because it has been in production for twenty years. This happens when it hinders business growth, when: every change requires weeks of analysis, knowledge is concentrated in the hands of just a few people, integrations depend on specific developments, or data remains trapped in architectures that are difficult to utilise. Furthermore, the energy transition, automation and digitalisation have brought new generation, production and operational capabilities into focus. However, in order to tackle these challenges, it is essential to modernise control and management platforms that were built decades ago.
A system becomes legacy when it hinders the business’s development, when every change requires weeks of analysis and when knowledge is concentrated in the hands of just a few people
In the industrial sector, this backlog – arising from systems that are technologically obsolete – has particularly far-reaching consequences, as, in addition to affecting the IT department, it limits the ability to connect plant systems with corporate systems (integrating MES and ERP, ensuring real-time traceability, implementing predictive maintenance, or using AI to analyse the data generated by machines, processes and products).
In line with this vision, the 'HFS Horizonsreport: Legacy Application Modernisation Services 2025’, based on a number of sources, identifies the following as the main impacts of legacy technology: higher operating costs (42 per cent), delays in digital transformation (38 per cent), an inability to scale (36 per cent) and difficulties in integrating modern platforms (30 per cent). 27 per cent also mention security vulnerabilities inherent in these environments
From migrating code to rebuilding knowledge
Traditionally, system migration began with the code, but in the case of AI-driven migrations, it begins with knowledge. The process should not begin by asking which technology we need to replace, but rather which aspects of our industrial know-how, our processes and our future capacity for innovation continue to depend on systems that are becoming increasingly difficult to understand and develop.
To this end, at Izertis, we rely on specialists who can scan through large codebases, analyse dependencies, link programmes to databases and interfaces, reconstruct functional flows, and compare actual behaviour with documentation that is often incomplete. Based on this understanding, they can draw up specifications, propose refactorings, transform components, write tests and maintain traceability between requirements, code and evidence.
We are no longer talking simply about a co-pilot assisting a developer, but about specialised agents capable of collaborating with one another. One agent analyses, another documents, another proposes the transformation, another generates evidence and another validates the result. Based on its initial experience, McKinsey estimates that modernisation timescales will be reduced by 40 per cent to 50 per cent, and that costs associated with technology debt will be cut by around 40 per cent.
Modernising to reshape the way the industry operates
It would be a mistake to use this new capability solely to convert COBOL to Java, update a framework or migrate an application to the cloud. That could breathe new life into the technology whilst retaining all the limitations of the original system.
Legacy modernisation should be approached as a comprehensive transformation process rather than as an isolated technology migration project
We must consider a broader opportunity focused on reviewing which processes still make sense, which rules should be retained, which features can be removed, and what architecture the industrial business needs in order to evolve. The system migration process must also involve decoupling domains, exposing capabilities via APIs, unlocking previously restricted data, integrating plant and corporate systems, and establishing a composable infrastructure on which to deploy new digital services.
At Izertis, we view this transformation as an end-to-end engineering process: assessment of the legacy system, preservation of functional knowledge, architectural redefinition, AI-assisted migration, specification-driven development, QA, validation and governed operation.
This vision is underpinned by a SDLC (Software Development Life Cycle) that has been adapted to incorporate AI across the entire software engineering cycle. At Izertis, we do not view AI as a tool that is used solely during the code-generation phase; rather, we incorporate it throughout the entire cycle, from the discovery and analysis of the existing system through to functional specification, architecture, development, testing, security, deployment and operation.
A particularly important element introduced in this migration cycle is SDD (Spec-Driven Development), in which the expected behaviour is specified and defined before code is generated; this specification serves as the reference for linking requirements, implementation, testing and evidence.
In the modernisation of legacysystems, where preserving decades of functional knowledge is critical, this approach becomes even more valuable.
Speed without rigour merely creates a new legacy
AI may misinterpret a rule, overlook a dependency, or generate a transformation that is technically valid but functionally incorrect. It also introduces risks relating to code and data exposure, dependence on suppliers, variability in results and the propagation of errors between agents.
That is why responsible modernisation requires functional equivalence testing, quality gates, security analysis, human oversight, observability and the ability to roll back. The approach adopted by Izertis involves incremental validation, SDD, specialist review and traceability between requirements, code, tests and evidence. It is not a question of regulating AI to slow it down, but of building the trust needed to scale it up.
Conclusion
For years, we have regarded legacy systems as a problem that needed to be solved. Agentic AI allows us to start seeing it in a different light: as an opportunity to recover knowledge, streamline processes, unlock data and rebuild digital capabilities.
For this reason, at Izertis we believe that legacy modernisation should be approached as a comprehensive transformation process rather than as an isolated technological migration project. The competitive edge will go to those industrial companies that use this capability to unlock their data, redesign their processes, connect the factory with the business, and build an industry that is ready to operate more intelligently, flexibly and autonomously.