Serguey Asael Shinder – IT Professional

For a while I was quietly frustrated with the model. I’d ask for help and get something generic, slightly off, not quite aware of my situation. I concluded it just wasn’t that good at the hard stuff.

Then I watched what I was actually feeding it. “This doesn’t work, how do I fix it?” No error message. No mention of the constraints I was working under. Nothing about the three approaches I’d already ruled out. I was asking a brilliant stranger to help with a problem I’d described in a sentence, and blaming the stranger for the vague answer.

The model can’t see my codebase. It doesn’t know my deadline, my platform, my weird legacy requirement. It knows exactly what I put in the prompt and nothing else. A starved prompt gets a starved answer, every single time.

So I changed my half of the conversation. Now I give it the actual error, the real constraints, what I tried and why it failed. And the quality of what comes back changed completely — because I’d finally stopped hiding the facts it needed.

I used to think using AI well was about clever prompting tricks. It isn’t. It’s mostly about being honest and complete about the context. The model supplies the fluency. My job is to supply the truth it can’t see.

– Serguey Asael Shinder