RELÈVE: Modernizing Your Legacy Systems with AI

Your legacy systems carry your most critical processes and represent years of investment. They work, and that is exactly why nobody dares touch them. Yet every year, the technical debt they accumulate weighs a little heavier.

Understand, transform, innovate, without ever rewriting blindly: that is the principle behind RELÈVE, PlanAxion’s AI-assisted application modernization offer. Here is what it changes, and where to start.

According to the Auditor General of Canada (Report 7, 2023), two thirds of the roughly 7,500 applications used by 45 federal departments and agencies are obsolete, and 562 of them are deemed essential to the health, safety or economic well-being of Canadians.

Why is modernizing your legacy systems becoming urgent?

Because five pressures add up every year: scarce skills, lost knowledge, cybersecurity exposure, rising costs and a brake on growth. They show up in almost every organization we meet.

• Scarce skills. COBOL, PL/I, Unibasic, PowerBuilder, VB6, Delphi: the experts are retiring and the market is not replacing them.
• Undocumented knowledge. Business rules live in the code and in the heads of a few people, and every departure is an irreversible loss.
• Cybersecurity exposure. Obsolescence is a risk in itself: end-of-life components that no longer receive patches, known vulnerabilities left open, outdated authentication and encryption.
• Rising costs and risks. Maintenance keeps getting more expensive, security is hard to guarantee, compliance is increasingly hard to demonstrate.
• A brake on growth. Integrations that cannot be built, data locked in, new markets out of reach: change is no longer an option, it is an obligation.

The cybersecurity angle deserves a pause, because it changes the nature of the urgency. As long as a platform is merely old, you can choose to wait.

Once it is no longer supported, waiting becomes a decision in itself: patches stop, published vulnerabilities remain exploitable, and the logs these systems produce fit poorly with current detection and monitoring tools.

The GAO (2019) analyzed 65 U.S. federal systems and singled out the ten most critical: they were 8 to 51 years old, and one of them carried 168 high or critical vulnerabilities.

Add to this the difficulty of demonstrating compliance with requirements that keep tightening. When nobody can explain what the code actually does, the simplest audit questions become impossible to answer with certainty: where is the sensitive data, who accesses it, under which rules. Québec’s Law 25 asks exactly those questions.

The risk does not stabilize; it grows with every year of waiting. Taking back control of that knowledge is as much a security measure as a modernization project.

Do you have to rewrite everything to modernize a legacy system?

No. Your systems work, and the business logic already exists: it is simply buried in the code. AI now makes it possible to turn your existing systems into specifications for what comes next.

You do not rewrite blindly: you extract, you validate, then you decide. AI speeds up the work; humans guarantee the quality. That is why RELÈVE combines three inseparable ingredients.

• Artificial intelligence. Accelerated specification extraction, documentation and code generation.
• Human expertise. Validation by your business experts and our architects at every phase.
• Structured governance. End-to-end traceability, auditability and compliance.

What does AI make possible from the source code?

Three levels of understanding, straight from the code, without depending on any original documentation. The analysis starts from the source code itself and climbs progressively toward business meaning.

• Semantic navigation. Mapping of modules, data dependencies and calls between programs.
• Intent extraction. What each routine actually does, from low-level functions to business logic, translated into plain language.
• Specification generation. Inputs, outputs, business rules and edge cases, structured and readable by your analysts and subject-matter experts.

More than COBOL: business logic is rarely concentrated in one place. It is spread across several technology layers, and our extraction accounts for that.

• Application code. COBOL, PL/I, Unibasic, PowerBuilder, Delphi, VB6, older .NET: the programs that carry the core business rules.
• Orchestration and transactions. JCL for batch, CICS for transactions, BMS screens: how everything runs and chains together.
• Data and storage. DB2, VSAM, IMS and proprietary databases: stored procedures, file structures and logic buried in the data layer.

To place these tools against classic approaches, read the difference between traditional AI and generative AI.

What does the RELÈVE offer include?

Three service streams that adjust to each organization’s maturity and application portfolio. They can be combined or deployed separately.

Core systems modernization

AI-driven specification extraction, validated by your experts. For each module, an explicit decision: retire, migrate to packaged software, or rebuild. Code generation is then AI-assisted and reviewed by our architects.

Integration and coexistence

Real-time replication, a data lake and COBOL routines wrapped as web services connect your applications without modifying existing structures. Coexistence between old and new stays under control throughout the transition.

Application portfolio management

Automated inventory and documentation of the portfolio, categorized by business domain, complexity, criticality and dependencies. From that overall view flow a multi-year roadmap and the governance that goes with it.

How does the RELÈVE approach unfold?

In two phases: first understand, then transform. No decision is made without specifications validated by your experts.

Understand: the RELÈVE Diagnostic

1. Extraction. AI analyzes the source code and produces detailed specifications per module, categorized by domain, complexity, criticality and dependencies.
2. Validation. Your business experts review and validate the specifications. Corrections, enrichments and category adjustments follow.
3. Decision. For each module: retire it, migrate to an existing system, or rebuild it.
4. Roadmap. A prioritized plan: schedule, risks, dependencies and procurement of the required systems and integrators.

Transform: execution

5. Generation. AI-accelerated code, reviewed by our experts.
6. Quality assurance. Quality, compliance, security and performance.
7. Deployment. Preparation and support for go-live.
8. Post-implementation. Knowledge transfer and operational continuity.

The solution architect’s role stays central at every one of these steps: AI proposes, the architect decides.

What does RELÈVE change for you?

Less risk, more speed, and knowledge that stays in the organization.

• Reduced risk. Every module is specified, validated and categorized before any decision. No blind cutover.
• Acceleration without compromise. AI absorbs the repetitive analysis and transposition; your experts and ours focus on the validations that matter.
• Preserved knowledge. Business rules buried in decades of code are recovered and documented in plain language, even without original documentation.
• Full traceability. From initial analysis to compliance testing, every decision and every generated line is auditable.

You do not rewrite blindly: you extract, you validate, then you decide.

What concrete evidence already exists?

The approach is not theoretical: here are excerpts of work PlanAxion has already delivered on existing systems.

• Unibasic legacy system. Reverse engineering of a system with no documentation at all, and full recovery of the business logic buried in decades-old code.
• Modern C# and .NET code. Extraction of functional specifications, decoupling of application logic from the platform, then a Windows-to-Android port carried out 100% by AI and reviewed by our architects.
• Typical deliverables. Specification book per module, state-transition specifications, dependency and call maps, application portfolio categorization.

Where do the AI tools run and what happens to your data?

Your data stays under your control: three deployment modes are possible, and the same governance and security framework applies in all three.

• Your sovereign environment. The AI tools run on your premises; code and data never leave your perimeter.
• Your approved AI environments. We work inside the platforms your organization has already authorized, for example dedicated cloud instances.
• A PlanAxion-managed environment. An isolated workspace, set up to your security and compliance criteria, to start quickly.

The RELÈVE governance framework covers regulatory compliance, control of sensitive data, full traceability of the work and auditability of the processes. Validations, automated and human alike, are adapted to your internal standards rather than imposed from outside.

Which platforms should a legacy system migrate to?

Each target is chosen with you, module by module, never imposed. The source technologies we meet most often are COBOL and PL/I on mainframe (JCL, CICS, BMS), Unibasic, PowerBuilder, Delphi and VB6, MS Access applications, older Java and .NET, and data held in DB2, VSAM, IMS or proprietary databases.

The modernization trajectories fall into a few families.

• Packaged software and existing solutions, when migration beats rebuilding.
• Modern Java, .NET, Python or TypeScript, when a custom rebuild is warranted.
• APIs, microservices and routines wrapped as web services, to open up the existing system without rewriting it.
• Cloud (Azure, AWS, GCP) or sovereign hosting, depending on your requirements.
• Real-time replication, data lake, and living documentation generated and then maintained.

Why entrust this modernization to PlanAxion?

The rigour of the large integrators, the agility of a human-scale firm, and the strength of an ecosystem. PlanAxion is a founding member of Axonal, a Québec leader in artificial intelligence that brings together three complementary areas of expertise: Baseline for artificial intelligence, Paragon Faction for technology and data management, and PlanAxion for process optimization and project management.

Together, the ecosystem counts more than 188 clients and more than 276 projects.

• Proximity and commitment. A leadership team involved on every engagement, fast decisions, a direct relationship.
• A pool of qualified experts. Our staffing practice gives us fast access to rare profiles: architects, analysts, specialists in older technologies.

This practice extends our AI Workshop, which helps organizations prioritize their AI use cases in four weeks. To compare the formats offered in Québec, read how to choose an AI workshop.

Where should you start: the RELÈVE Diagnostic?

With a progressive, concrete approach, over a few weeks, at a fixed price and with no commitment to what follows. The results speak for themselves, whether you continue with us or not.

1. Choose 2 to 3 representative modules, of varied complexity and nature, to test the approach under real conditions.
2. Run the extraction and evaluate: within a few weeks, concrete specifications your business experts can judge directly.
3. Map the architecture and master data: inventory of the technologies in place, migration patterns and supporting systems.
4. Decide on concrete grounds: you then have everything you need to plan what comes next with confidence.

You leave with:

• validated specifications for the pilot modules;
• a portfolio categorization by domain, complexity and criticality;
• an architecture and migration-pattern map;
• one decision per module: retire, migrate or rebuild;
• a prioritized roadmap and a budget order of magnitude.

Let’s pass the torch together. What follows comes in three steps: a 45-minute meeting to understand your application portfolio and priorities; a RELÈVE Diagnostic on 2 to 3 pilot modules, with extraction, validation and mapping; then an informed decision, where you choose the trajectory, with us or with the deliverables in hand. Book a 45-minute meeting.

Frequently asked questions

What is a legacy system?

A legacy system is an application still in production but built on older technologies such as COBOL, PL/I, Unibasic, PowerBuilder, VB6, Delphi, MS Access, or older Java and .NET. It often carries critical processes, costs a lot to maintain, and depends on skills that are increasingly rare on the market.

Can AI really read COBOL code without documentation?

Yes. The analysis starts from the source code itself: the models map modules and dependencies, then describe in plain language what each routine does. The resulting specifications must then be validated by your business experts, which remains the decisive step before any transformation is attempted.

Do you have to rewrite everything to modernize a legacy system?

No. The decision is made module by module: retire what is no longer used, migrate to packaged software when it covers the need, rebuild only where justified. Routines can also be wrapped as web services to open up the existing system without rewriting it.

How long does the RELÈVE Diagnostic take?

A few weeks, on 2 to 3 representative pilot modules, at a fixed price with no commitment to what follows. You leave with validated specifications, a portfolio categorization, an architecture map, one decision per module and a prioritized roadmap with a budget order of magnitude.

Where do the AI tools run and what happens to our data?

You choose one of three modes: your sovereign environment, your already-approved AI platforms, or an isolated environment managed by PlanAxion. In every case, code and data stay within the agreed perimeter, and the same governance, traceability and auditability framework applies throughout the engagement.