On AI Tools
Last week, I built a full-stack application in an hour using Lovable.dev and Cursor. Database, backend, polished UI. I felt like unstoppable.
Then I tried to add user authentication. The AI-generated code had security holes you could drive a truck through. The database schema fell apart when I needed to add a related table. The "polished" UI broke and it wasn’t easy to fix.
In fact, I spent the next three days fixing what took one hour to build.
This is the reality of AI tools that nobody wants to discuss: They're incredible for prototypes and tricky for product development.
What AI Tools Actually Do
I've been using Cursor, Loveable.dev, and ChatGPT daily for months. Here's what I've learned:
They make easy things trivial. Building a contact form? Five minutes. Creating a CRUD app? Twenty minutes. These used to take hours.
They make medium things dangerous. Complex business logic? The AI will confidently generate code that looks right but fails edge cases you haven't thought of.
They make hard things impossible. System architecture? Performance optimization? Security? The AI will hallucinate solutions that sound plausible and fail spectacularly.
The Real Divide
The divide isn't between who’s using AI and who isn’t. It's between those who understand what AI can and can't do, and those who don't.
I watched a junior developer build an AI-generated code that leaked user data. He was curious. He was exploring. He just didn't know enough to recognize bad code when the AI produced it.
I watched a senior engineer use AI to prototype six different approaches in an afternoon, then build the actual solution by hand. She knew exactly where AI helped and where it hindered.
The knowledge still matters. You just apply it differently.
How Roles Are Actually Changing
Engineers: AI helps with boilerplate but can't architect systems. The engineers thriving are those who use AI for the boring parts and focus on the hard parts.
Product Managers: AI can summarize user research but can't make product decisions. PMs who survive will be those who understand AI's outputs are starting points, not answers.
Designers: AI can generate variations but can't understand context. Designers who last will be those who know why something feels wrong, even when the AI insists it's right.
The Uncomfortable Truth
Yes, AI tools are revolutionary. Yes, they're changing how we work. But the breathless evangelism misses crucial points:
AI amplifies existing skills; it doesn't replace the need for them
Quality still requires expertise to recognize and achieve
Most AI-generated work is mediocre (that might be enough for many use cases)
The real skill is knowing when to use AI and when not to
A More Honest Choice
You don't have to choose between embracing or rejecting AI. The real choice is: Will you understand these tools' limitations and use them wisely? Or will you abdicate judgment to machines that have no concept of quality, correctness, or consequences?
I still use AI tools every day. They've made me more productive. But they've also shown me that expertise matters more than ever.
Because now everyone can generate code. But only those who understand it can tell if it's any good.