Since launching EigenStrategy, I have worked directly with buy-side credit teams on AI implementation. The gap between what the headlines suggest and what actually happens inside these firms is significant. Three things account for most of it.
Data security and compliance
Sending sensitive analysis, position data, or anything MNPI-adjacent through a cloud API is still not acceptable at many firms without the right controls in place. The compliance question comes first, before architecture, before model selection, before anything else. For a long time, there was no clean answer to it.
Local and controlled deployment has only recently become a viable option at meaningful model quality. That shift is real, but most institutions are still working through what it means for their stack. The firms that waited for a compliant path are now starting from scratch while others are a year ahead.
Infrastructure design
In practice, architecture matters more than model upgrades once the basics are in place. How context is structured, retrieved, and routed through a workflow has a larger impact on output quality than most firms expect when they start.
The common failure pattern: a team gets access to a strong model, runs some promising tests, and then tries to plug it into existing workflows without changing how information flows. The output is inconsistent. The team concludes the model is not good enough. The actual problem is that the surrounding system was not designed for the task. A stronger model sitting on top of a weak information architecture produces unreliable results at scale.
Cultural change
Training is easy to sell. Implementation asks people to change daily habits, which requires organizational commitment most institutions are not yet ready to make.
The firms making real progress are usually not the ones talking about it most publicly. Working advantages in markets tend to stay quiet. The teams with genuine adoption have a senior champion, a specific workflow they redesigned around the tool, and a clear measure of what better looks like. Those conditions are less common than the conference circuit suggests.
What comes next
Within the next six to twelve months, there will be a visible performance gap between firms that embedded AI into actual workflow and firms that did not. Investment process speed, research coverage depth, cost per analysis. The gap will show up in the numbers before it shows up in the narrative.
The bifurcation is already underway. Most firms will not see it clearly until it is too late to close quickly.