AI Engineer World's Fair 2026 — unofficial import demo

A couple of years ago, everyone worried about AI hallucinating. We rarely hear that word anymore, but it’s just because the problem grew up. Today, your AI still doesn’t know how to say “I’m not sure.” Instead, it hands you a revenue number that’s wrong in ways that look exactly like being right.

The good news is we already solved this once, for people: you onboard a new hire so they understand your business; you put subjective, high-stakes calls in front of more than one set of eyes. This talk walks through patterns we run at Upside, including a librarian every agent consults before it acts, a jury-and-judge model for the questions a single pass can’t be trusted to answer, and knowing when the model itself is just too dumb for the job. Live demos and real failures included.