Over the past few months I have been watching less for improvements in the models themselves and more for signs of institutional adoption. This week brought another one: Millennium and Anthropic announced they are building a digital risk analyst inside Millennium's AI lab, using Claude to explain changes in portfolio risk alongside human risk managers.
The interesting part isn't the technology. It's that more investment organizations are starting to treat foundation models as infrastructure rather than experiments. The implementation paths will vary by culture as much as by capability: how decisions get made, how much authority people are willing to hand to a system they didn't build themselves.
The observation I keep coming back to is that the hardest part isn't picking the model. It's the bridge in the middle, what I'd call a forward deployed role inside an investment firm: translating how a firm actually makes decisions, its processes, heuristics, and accumulated judgment, into something AI can use, and translating back what the technology can and cannot do.
Models will continue to improve. The organizations that benefit most won't necessarily be the ones with access to the best models. They'll be the ones best at translating decades of institutional knowledge into systems that can actually use it.