The Risks We Introduce When AI Writes Code

The AI revolution in software development is real. But the code that works today might collapse tomorrow. What development teams don't know (yet)

From Shovel to Automated System

In recent years, we've witnessed a radical transformation in how we write software. If it once felt like working manually with a shovel, today it's like having a fleet of automated combines running on their own. But here's the detail that immediately catches CFOs' attention: an AI harvester costs significantly less than an actual developer, even a junior one.

AI has made software development democratic, fast, and above all accessible. The numbers are undeniable: development times shrink, costs drop (at least in appearance), and now even non-developers can write working code. The gap between a junior developer and a non-developer? It's practically disappeared. AI has absorbed all the foundational knowledge of a junior developer—and in many cases, surpasses it.

As a result, companies have grasped one simple truth: they need smaller teams. Fewer developers, same (or larger) deliverables. The harvesters run on their own. Why keep expensive agronomists if the machine already knows the trade?

The Real Problem: AI is Nearsighted by Design

But here's where the trap hides.

AI is trained for one thing only: reaching the objective you set for it, using the fewest tokens possible. It cannot—and should not—account for all the architectural aspects of software development. Cybersecurity, the principle of least privilege in access controls, database choices, proper data storage: these are global context problems that require strategic vision.

AI was trained to solve the immediate problem. Not the long-term vision of your business.

What happens then? AI deprioritizes cybersecurity. It ignores the principle of least privilege. It chooses to store data in the wrong locations—maybe locally when it should be in the cloud, or vice versa. It doesn't have a long enough vision to choose between monolithic and microservices architectures, between a single frontend and microfrontends. It doesn't know the right deployment strategy based on your team's actual capabilities—because maybe your team doesn't know how to deploy on the cloud, but AI suggests it anyway.

And Then Growth Arrives

The first 12 months go wonderfully. The team is fast, deliverables are on schedule, the budget is under control. It feels like magic.

But as your project grows, the context becomes increasingly large. And you need ever more tokens to manage that context—the cost of AI stops being negligible and becomes a significant line item in your budget. AI's performance naturally degrades—just like a human's would when managing too many things simultaneously. Generated code becomes increasingly fragmented. Integrations start to conflict with each other. Security proves inadequate. Cloud costs explode because AI never thought about global optimization.

This is the moment of truth: the code that worked perfectly in year one now generates technical debt at unsustainable rates. Your fleet of autonomous harvesters reveals itself for what it is: a tactical tool, not a strategic one.

Tools Will Arrive. But They Don't Solve the Real Problem.

My vision is that in the coming years we will see a proliferation of new tools designed exactly for this: guiding AI. Standardized procedures. Pre-packaged architectural patterns. Frameworks that force the model to respect constraints around security, scalability, and governance.

They'll work. Up to a point.

Because here's another truth that nobody wants to say out loud: when a company scales, nothing stays standard. Every company is unique. Its needs evolve. Its constraints are particular. Its strategic vision never matches another organization's.

And in that moment—when standardization stops working—humans make the difference.

The Solution: Invest in People, Not Technology

The company that will succeed tomorrow isn't the one with more AI tokens at its disposal. It's the one that has people capable of managing AI and making it follow the company's vision.

These people must be different from traditional developers. They must be:

• Informed — constantly updated on the latest technological developments
• Problem solvers — capable of looking at a problem from a strategic perspective, not just tactical
• Decision architects — able to choose which technology, which pattern, which approach is right for the specific context
• People managers — capable of guiding heterogeneous teams (developers, non-developers, AI tools, technical stakeholders)

In other words: they must be junior IT managers. Guided by a CTO with an even broader vision.

The Role of the CTO in the AI Era

Tomorrow's CTO doesn't write code. Or at least, not for 90% of their time. Tomorrow's CTO:

• Defines architectural standards that AI can follow (and that the team understands)
• Creates governance for AI-generated code—because yes, automatically generated code needs oversight
• Chooses which AI to use, when to use it, and where to draw the line
• Anticipates where technical debt will accumulate and how to manage it
• Builds teams that understand the "why" behind technical decisions
• Translates strategic vision into technical principles

The CTO is no longer a technical expert who knows everything. They're a strategist who knows how to choose, guide, and anticipate consequences.

What to Do Right Now

If you're a company today, your options are two:

• Option A: Let yourself be guided blindly by AI. It will work in year one. Then you'll need to hire senior developers to clean up the mess—at triple the cost.
• Option B: Hire (or develop internally) a CTO who knows how to manage this transition. Invest in technical governance now. Sleep well for years to come.

The choice is yours. But the companies that make the right choice today will beat those that wait until tomorrow.