Morgan Stanley wealth unit moves AI to agentic workflows

Morgan Stanley's current position rests on a data decision made roughly eight years ago. The wealth division initially thought about data account by account, but clients wanted a view of the entire relationship. That reframing led to machine learning capabilities such as next-best action, which gives advisers guidance on next steps based on the broader portfolio of data. Shamsi traces today's agentic systems back to that foundational work, noting that agents can only be as good as the context and data behind them.

The division's three focus areas — client interaction, service and adviser assist — reflect a deliberate division of labor. The service assist agent handles operations that span multiple applications, while the adviser assist agent aims to remove friction from day-to-day functions so advisers can spend more time on client engagement. Human decision-making stays in the middle by design.

As capabilities expand, Shamsi's emphasis on controls suggests the constraint on agent autonomy is intentional rather than temporary. The agents operate within defined boundaries on data access, permitted actions and when a human must intervene — a model other firms deploying similar workflows are likely to watch.