I had a conversation with a board member recently that I think a lot of technology leaders are having right now. We were discussing the budget for a legacy migration project—moving a decade-old monolithic application to a microservices architecture.
He looked at the timeline, then at the headcount, and asked the question that has become the elephant in every boardroom:
“Can’t we just feed the codebase into an AI and ask it to rewrite this in Python? Shouldn’t this cost, like, $20 a month?”
It’s a valid question for someone who sees the viral Twitter threads of people building Flappy Bird clones in 30 seconds. If AI can write code, why are we still paying for a full engineering department?
Here is the hard truth that I need every stakeholder to understand: AI can build you a toy for free. But an enterprise engine costs money to build, and even more to keep running.
The “Hello World” Trap
There is a fundamental misunderstanding about what “coding” actually is in an enterprise context.
If you ask a sophisticated LLM to “create a To-Do list app with a login screen,” it will do a phenomenal job. It will give you the HTML, the CSS, and the React components. It might even spin up a simple database schema. It feels like magic. It feels free.
But that creates a dangerous illusion of competence. It’s the difference between building a doghouse and building a skyscraper.
If I give you a pile of lumber and a hammer, you can probably build a doghouse. If I give you AI, it’s like having a robot that nails the wood together for you. But if you try to build a 40-story skyscraper using that same methodology, the building will collapse the moment the wind blows.
What AI Misses (And Why You Need a Team)
When we talk about “Enterprise Application Modernization” or “Migration,” we aren’t just typing syntax. We are dealing with an ecosystem. Here is why a prompt isn’t going to replace my architects and DevOps engineers anytime soon.
- Context is King (and AI doesn’t have it) Your legacy code isn’t just “old code.” It is a historical record of ten years of business decisions, regulatory requirements, and weird edge cases. AI looks at a block of spaghetti code and translates it literally. A human engineer looks at it and asks, “Why does this function pause for 200ms? Oh, it’s because of a race condition with the legacy payment gateway.” If you auto-migrate that without understanding the why, you crash production.
- The Non-Functional Requirements You can’t prompt-engineer “security at scale.” You can’t ask ChatGPT to “make it GDPR compliant” and hope for the best. Enterprise apps live or die by their non-functional requirements: latency, throughput, security, observability, and resilience. Designing a system that handles 100 concurrent users is free; designing one that handles 100,000 without leaking data requires human intent and architectural rigour.
- Integration Hell No enterprise app lives in a vacuum. We are connecting to CRMs, ERPs, third-party APIs, and obscure banking protocols. AI struggles immensely when it has to debug a connection between a modern cloud function and a mainframe that speaks a dialect of COBOL from 1988. You need a human to bridge that gap.
The Real Role of AI: The Ultimate Force Multiplier
Now, don’t get me wrong. I am not anti-AI. In fact, I mandate the use of AI tools for my teams. But we need to reframe how we view the cost savings.
AI doesn’t make the project cheap; it makes the project possible within a reasonable timeframe.
In the old days, a modernization project might take 18 months. With AI handling the boilerplate code, generating unit tests, and explaining complex documentation, maybe we can do it in 9 months.
- We aren’t firing the architects; we are freeing them from writing getters and setters so they can focus on system design.
- We aren’t firing the QA team; we are using AI to generate edge-case scenarios that the humans then verify.
The “Technical Debt” of Cheap
There is an old saying in engineering: Cheap, Fast, Good. Pick two.
With AI, people think they have found a cheat code to get all three. But if you rely solely on AI to build your critical infrastructure to save money, you are racking up Technical Debt at an alarming rate.
Code generated by AI without heavy human oversight is often verbose, insecure, or hallucinatory. If you fire your team and let the “bot” build the app, you won’t pay in salaries today. But you will pay ten times that amount next year when the system creates a security vulnerability that costs you your reputation, or when you need to add a feature and realize nobody knows how the AI-written code actually works.
The Verdict
AI is not a replacement for a carpenter; it is a power drill.
Before power tools, it took a long time to build a table. With power tools, it’s faster. But you still need a master craftsman to measure, cut, and ensure the table doesn’t wobble when you put dinner on it.
For our stakeholders and clients: Do not expect a $0 price tag on your next migration just because ChatGPT exists. Expect a more efficient, higher-quality delivery from a team that is smart enough to use AI as a tool, not a crutch.
We are still the pilots. AI is just a really, really good engine.