There’s a specific kind of cognitive dissonance that comes with being a DBAI student studying AI while also being someone who uses AI tools as a core part of how I work. I spend mornings reading papers about the risks of autonomous agents. I spend afternoons asking autonomous agents to help me organise my notes. The gap between the research and the lived experience is both fascinating and a little disorienting.
The thing that strikes me most is how differently AI behaves when you’re paying close attention versus when you’re just using it as a tool. When I’m in “work mode,” I accept AI outputs as basically reliable. When I’m in “research mode,” reading about hallucination rates and coordination failures and agents that report task completion when they’ve done nothing of the sort — I become much more suspicious of the exact same tools.
I don’t think either mode is wrong. But sitting with both at once makes you realise something: most people who use AI daily have never been exposed to the research on how AI fails. And most researchers studying AI failure haven’t used it as a daily work tool at scale. There’s a huge gap between those two groups, and it’s producing blind spots on both sides.
That gap is actually what I want to close with this blog — not with alarmism, not with hype, but with honest translation between what the papers say and what it actually feels like on the ground. That feels worth doing.