The AI Code Illusion: Why 300% More Code Doesn't Mean Better Software

Over the last two years, the software development industry has been obsessed with a single metric, speed.

With the explosion of AI coding assistants, standard web developers and junior engineers are churning out applications at a pace that was unimaginable just a few years ago. But a new MIT research study highlighted today by the Financial Times exposes the dark side of this AI-driven gold rush.

The data reveals a massive, fundamental disconnect between generating code and actually building digital infrastructure that works.

The Productivity Mirage

The numbers in the report are staggering, but they tell a cautionary tale for any business investing in digital tools.

When tracking developer output using AI, researchers found an explosive impact at the top of the funnel, coders created or edited almost 300% more files. On paper, this looks like a revolution. An agency can now build your new portal or internal dashboard three times as fast.

But then, the funnel collapses.

• That 300% boost was halved to a 150% increase when measuring discrete pieces of work submitted for review.
• It then shrunk drastically to a mere 30% uplift when measuring the actual production of a company’s core product.

The takeaway is brutally clear, AI is making it incredibly easy to generate raw code, but it is not making it any easier to build valuable software.

More Code = More Technical Debt

If your developer is using AI to generate 300% more files, but your business is only seeing a fraction of that in actual operational value, what is happening to all that extra code?

It is becoming technical debt.

When you hire a standard web agency to build a custom application, and they rely heavily on AI to just “spit out” the code, you are paying for bloat. You are getting heavy, unoptimised, and often fragile codebases. It might work on launch day, but six months from now, when you need to integrate your accounting software or scale your user base, that tangled web of AI-generated files becomes a nightmare to untangle.

This friction is hitting the enterprise level, too. Uber’s CEO, Dara Khosrowshahi, recently revealed that the company blew through its entire AI budget in a single quarter, forcing them to pivot back to lower-cost models. If the giants are struggling to find efficient ROI from raw AI usage, local businesses shouldn’t be paying agencies to experiment with their infrastructure.

The Architect’s Advantage

This is exactly why I don’t sell “websites” or standard development hours. I sell Architecture-as-a-Service.

When I build custom digital infrastructure for businesses in Cheshire and the Wirral, I don’t use AI to indiscriminately bloat the codebase. True engineering is about subtraction, not addition. It is about understanding the exact operational bottleneck a business is facing and designing the absolute leanest, most secure, and fastest system to solve it.

AI is an incredible tool, but it is not an architect. It cannot understand your specific business logic, and it cannot plan for your three year scaling strategy.

If your agency is bragging about how fast they can code your next project, you are asking the wrong questions. The question isn’t how much code they can write. The question is how much code they can elegantly eliminate.