AI Efficiency in Research Ops

It feels like media and conversation around AI swing between two extremes:
“It will replace us” and then "It isn’t reliable enough to be useful.”

In practice, AI seems to work best as a thinking partner and early workflow accelerator, not an autonomous agent (not yet). At least, this was my and my team's experience over the past year, every month or so we would touch base to share notes on what was working.

I mapped my reflections on how we used Co-pilot's integration of GPT 5 in the research function of a digital bank below.