Serguey Asael Shinder – IT Professional

I spent an embarrassing stretch of time being frustrated with a model for giving me generic answers, before I realized the answers were generic because my questions were. I was asking it to solve problems while withholding almost everything it would need to solve them well, and then blaming the tool for guessing.

The turning point was a small one. I’d asked for help with a piece of code and gotten back something reasonable-looking that didn’t fit my situation at all — wrong version, wrong assumptions, wrong constraints. I was about to dismiss it when I looked at what I’d actually typed. I had described the problem the way it lived in my head, where all the context was obvious, and forgotten that none of that context had made it into the words on the screen.

So I tried again. This time I said the stack and the versions. I said what I’d already attempted and why it hadn’t worked. I named the one constraint I couldn’t break. The same tool, given the same kind of problem, produced something genuinely useful — because this time it wasn’t guessing which of a dozen valid answers I needed. I’d told it.

That reframed the whole relationship for me. The model can’t see my codebase. It doesn’t know my deadlines or my history with a problem. It has no access to the enormous, invisible context I carry without noticing. When I leave that out, it fills the gap with the average answer, and the average answer is almost never mine.

Now I think of it like briefing a sharp new colleague on their first hour — capable, fast, willing, and completely blind to everything I haven’t said out loud. The quality of what I get back is mostly decided before the model does anything, in the paragraph of context I either bother to write or lazily skip.

The tool was never the bottleneck. The bottleneck was everything I knew and didn’t say. Most of the bad answers I ever got, I’d quietly asked for.

– Serguey Asael Shinder