Recently at AI Day 2026, Nandan Nilekani made a blunt statement:
Legacy modernization can’t be deferred anymore. Enterprises must clean up the mess of years.
He is absolutely right.
And I say this not as a commentator — but as someone who has lived through multiple waves of “modernization.”
The Old Days: VB6 to .NET Was Not a Transformation
Early in my career, while working with large enterprise customers through Infosys and later LTIMindtree (then L&T Infotech), I led multi-million-dollar legacy modernization initiatives.
Most of them were:
VB6 → .NET
Classic ASP → ASP.NET
On paper, we had “automated migration tools.”
Reality?
They mostly:
Converted syntax.
Broke business logic.
Didn’t ensure compilation.
Had zero guarantee of functional equivalence.
We were lucky if 50% of the system behaved correctly after tool-based migration.
I clearly remember late nights:
Fixing broken workflows.
Debugging edge cases no one documented.
Reverse-engineering business rules from 15-year-old code.
Struggling with zero test coverage.
The biggest problem wasn’t syntax migration.
It was extracting business intent from messy systems.
Fast Forward: AI Changed the Equation
Recently, I ran a serious PoC.
150+ API endpoints written in PHP → migrated to .NET Core.
Single-handedly. In 4 working days.
Let that sink in.
The validation test?
We pointed:
Existing web app
Existing mobile app
…to the new migrated APIs.
And it worked for ~90% of the cases immediately. ( This might not be the case for complex system but it would be much better than earlier tools for sure)
Not theory. Not demo. Real system.
Why AI Worked Where Old Tools Failed
The difference is fundamental.
Old migration tools:
Were syntax translators.
Had no reasoning ability.
No contextual understanding.
No business rule inference.
Modern AI models:
Understand patterns.
Infer implicit rules.
Identify structural gaps.
Suggest missing validations.
Generate integration test suites.
But here’s the key — and this is important:
AI works only when you:
Provide proper context.
Define project structure.
Supply boilerplate and architectural direction.
Upgrade context regularly.
Work in small vertical slices.
Continuously validate.
If you dump a 20-year-old monolith into AI and expect magic — you’ll fail.
If you guide it architecturally — it becomes a force multiplier.
The Real Breakthrough: Parallel Functional Equivalence
In the past, validating functional equivalence was a nightmare because:
No test cases
No documentation
SMEs had moved on
Edge cases lived in production only
Today, AI can:
Analyze old logic
Generate integration test scenarios
Suggest edge cases
Create regression test suites
That alone changes the economics of modernization.
The Harsh Truth
Legacy modernization feels scary because:
Systems are old.
Documentation is poor.
Teams lack domain memory.
Ecosystems are complex.
But with AI evolving at this pace, incremental rewrite and structured upgrade will become the new normal.
Enterprises that delay will not just face technical debt —
they will face competitive irrelevance.
This time, we don’t have the excuse of “tools are not ready.”
They are.
The question is:
Are architects and stakeholders ready to embrace disciplined, AI-assisted modernization instead of hiding behind risk?
I can speak on this for hours — especially from my experience leading modernization in insurance and life sciences domains.
But I’ll leave you with this:
Modernization is no longer a tooling problem.
It’s a leadership decision.
What’s your experience with AI in legacy transformation?
First published on LinkedIn.