I had a realization recently, dropping my car at the garage.
The mechanic explained the problem: faulty sensor, clever workaround on the wiring, partial rewire of the electrical harness. Technical, specific, the kind of thing that requires expertise and tools I don’t have.
I nodded along, grasping the “what” and “why,” but knowing I couldn’t pick up the tools myself even if I wanted to.
That’s exactly how I feel watching GenAI work on legacy RPG and COBOL systems.
The Parallel
GenAI acts like that expert mechanic:
It’s remarkably smart. It understands dense technical debt instantly. It spots issues I’d miss, improvises reliable workarounds when something fails, and does all of it far faster than I could manually.
According to Thoughtworks’ 2026 Looking Glass report on AI and software delivery, AI is powering “core renewal” by handling the heavy lifting of refactoring and integration—work that used to take months now takes days.
The result? We’re moving from “the person writing every line” to “the informed supervisor.”
We understand the diagnosis. We approve the strategy. We validate the work. But the actual execution—the heavy lifting—that’s the AI’s job now.
What This Changes
The shift from “I code” to “I oversee” is bigger than it sounds. Before, a developer spent 60% of their time on routine refactoring, boilerplate, and mechanical conversions. GenAI handles that now. The developer spends 60% of their time on judgment calls: does this business logic actually work this way? What edge cases exist? How do we validate correctness? That’s better work—more human work.
We’re seeing real-world data on productivity gains at enterprise scale. Refactoring projects that took 3–4 months manually now take 6–8 weeks with AI. Feature development speed increases 50–70%. Code review cycles compress because there’s less mechanical review and more semantic review. For IBM i shops, this is particularly valuable because the goal is often to modernize legacy applications while maintaining 100% business logic compatibility. GenAI accelerates the mechanics (converting SYNON to RPG Free, COBOL to modern COBOL) while humans focus on correctness—did we preserve the behavior that matters?
RPG developers command $100k–$120k+ salaries and the talent shortage is real. But with AI handling routine work, one senior developer plus GenAI can accomplish what used to require two senior developers. That changes your hiring problem economics. You still need experienced people, but you need fewer of them, and they spend their time on higher-value work. Technical debt that consumed 80–90% of IT budgets suddenly becomes addressable when you have the mechanical labor capacity. You can actually keep systems current instead of letting them slide into crisis.
What Doesn’t Change
Here’s what’s important: GenAI doesn’t replace judgment. It doesn’t eliminate the need for experienced people. It doesn’t remove human responsibility.
You still need developers who understand:
- Your business logic and why it exists
- Your integration points and dependencies
- Your performance constraints and why they matter
- Your data model and what it represents
GenAI is that expert mechanic. But you’re still responsible for understanding the work and validating that it’s correct.
If you hand GenAI a vague request and don’t review the output, you’ll get bad results. If you hand GenAI a precise, well-scoped request and review carefully, you’ll get excellent results 3x faster than before.
The Organizational Shift
The companies that are winning with AI for legacy modernization share a pattern:
They’ve accepted that this changes roles. Senior developers spend less time writing code and more time reviewing GenAI output, validating business logic, and making judgment calls. The team gets smaller but more strategic. The work happens faster.
The companies struggling with AI have still structured it as “replace developers” or “eliminate the modernization cost entirely.” When you do that, you get bad results because GenAI really does need human judgment.
Think of it like this:
Bad approach: “Use AI to fully automate our modernization project. We don’t need developers anymore.” Result: AI generates code that compiles but doesn’t preserve business logic. You discover this in production. Crisis.
Good approach: “Use AI to handle routine refactoring and boilerplate. Our senior developers focus on validating business logic and making judgment calls.” Result: Modernization happens 3x faster, quality is high, developers have more satisfying work.
A Curious Question
I’ll be honest: I’m watching a future scenario that nobody’s talking about yet.
What happens when GenAI is so productive that we don’t need as many developers, but we also don’t have people trained to oversee the AI? When the mechanic is so good that we forget how cars work?
Right now, senior developers are perfectly positioned to oversee GenAI because they actually understand the work. But that generation is retiring. The next generation has learned alongside GenAI. Will they have the context to know when AI is making the wrong choice?
I don’t know. We’ll figure it out. We always do.
For now, the immediate reality is: GenAI is like that expert mechanic. It’s making us more productive. It’s shifting our role from “doer” to “overseer.” It’s relieving pressure on our talent shortage.
And that’s genuinely good news for legacy systems.
If you’re thinking about how GenAI fits into your legacy modernization strategy—and specifically how to structure your team to use it effectively—let’s talk. We can help you think through how to position your developers and validate your approach before committing to a major transformation.