Hey everyone,
I wanted to share something that might be helpful—especially for folks who aren’t technical but are experimenting with AI coding tools. Consider this a little PSA.
Take a look at the attached screenshot. This is a real interaction I had with Claude:
-
At first, it claimed the problem was with how tokens (like passwords) were being hashed, and it generated a big diff touching 40+ lines of code.
-
The issue? That diagnosis was wrong. None of those changes were needed.
-
Once I pointed out the actual problem (that the token wasn’t being passed into the function at all), Claude corrected itself—and the real fix was just 1 line of code.
Here’s the challenge:
-
AI tools sound confident, even when they’re wrong.
-
Without a technical background, it’s easy to accept that first “solution” and accidentally introduce dozens of unnecessary changes.
-
Often, the real fix is far simpler—but spotting it requires human judgment.
An even bigger concern is when you’re using the agent mode. That’s more of a YOLO approach: it charges ahead making sweeping code changes, and if you’re not intentional about reviewing each diff, you can get into trouble quickly.
Don’t get me wrong—I use AI daily and it’s incredibly useful. But for anyone non-technical, this is the reality: sometimes the AI writes you a whole novel when the answer is a single sentence. And while agents can make things look effortless, the truth is that coding isn’t magically “solved” yet. Until AI gets much more efficient (and safer), the human element is very much still required—and in my view, always will be to some extent.
