AI cannot generate judgment

Do you know the Dunning-Kruger effect? The idea is that people with limited knowledge in a subject overestimate their competence because they don't yet know what they don't know. As they gain experience, they begin to realize how much there is to learn, and their confidence becomes more grounded.

While watching developers embrace AI agents over the past few months, I am wondering whether we are witnessing the same phenomenon. Since AI has made it easy to produce work that looks like it came from years of experience. AI has reduced the gap between appearing competent and actually being competent. And I think that's one of the most interesting challenges our industry will face over the following years.

AI has changed the learning curve

Years ago, if a developer wanted to build a distributed system, they had to spend months learning messaging, databases, networking, deployment, monitoring, and failure modes. The learning curve slowed them down. It forced them to understand the technology before they could build something meaningful.

Today, an AI agent can generate a microservice, a Kafka consumer, Kubernetes manifests, CI/CD pipelines, and infrastructure code in minutes. In many cases, a developer can assemble an impressive-looking system without fully understanding the technologies behind it.

The code runs, and confidence rises. But generating a system and understanding a system are two very different things.

Production is where understanding shows

The real difference appears months later, when the system is under real pressure.

A downstream dependency becomes slow. A queue starts growing. A service becomes unavailable. A retry storm begins. Suddenly, everyone discovers the difference between generating a solution and understanding how it behaves.

I have always believed that experience is what you got when you didn't get what you wanted. AI can implement patterns much faster, but it cannot replace the lessons learned from watching those patterns fail in production. That's where judgment is built.

The take

The people who thrive in the coming years are the people who can look at ten AI-generated solutions and calmly say, "Nine of these will hurt us six months from now. Let's do the boring one."

That's not coding. That's judgment. And judgment is usually by making mistakes, seeing systems fail, and living with the consequences long enough to recognize the patterns.