The Future of Systems: Why AI Is a Problem Disguised as a Solution

The Trust Problem#

More than ever, organizations are trusting AI to write, review, and ship code. But trust without verification is not confidence - it is recklessness. AI-generated code is not measurable in the same way human-written code is. When a developer writes code, there is a reasoning trail: decisions made, trade-offs considered, edge cases handled deliberately. AI produces output that looks correct. Looking correct and being correct are not the same thing.

This matters most where it is least visible - in critical infrastructure. The banking systems, healthcare platforms, energy grids, and transportation networks that society depends on. These systems cannot afford the "good enough" that AI optimization promises. When organizations cut costs by replacing careful engineering with AI-assisted shortcuts, they are borrowing against a future failure they cannot predict.

The Cheatsheet Effect#

For developers, AI has become a cheatsheet. And like any cheatsheet, it undermines the learning process it claims to accelerate.

Software development has always been an everyday discipline of learning. You read documentation. You debug. You understand why something works, not just that it works. This process is not inefficiency - it is how developers build the judgment that prevents catastrophic mistakes.

When we skip that process, we produce developers who can prompt but cannot reason. Who can generate but cannot debug. Who can ship but cannot maintain. The short-term productivity gain creates a long-term competence crisis.

What We Actually Need#

Instead of deeper AI adoption, I believe we need the opposite: more discipline, more standards, more rigor.

Specifically:

• Establish clear standards for how applications are built, checked, and tested. Not guidelines - standards. Enforceable, measurable criteria that every system must meet before deployment.

• Consolidate around fewer, proven languages. The fragmentation of the development ecosystem makes quality control nearly impossible. Choosing a smaller set of efficient, well-understood languages would make systems more auditable, more maintainable, and more secure.

• Treat AI adoption the same way we treat any other dependency - with skepticism. The same urgency we bring to adopting AI, we need to bring to evaluating whether it belongs in a given context. Some systems should never touch AI-generated code.

The Path Forward#

I am not against AI as a concept. I am against the way we are adopting it: uncritically, without measurement, and at the expense of the fundamentals that make software reliable.

The future of software development is not about replacing developers. It is about raising the bar for what we expect from the systems we build. That starts with standards, not shortcuts.