Bank of America is moving AI agents closer to the conversations where financial advice is prepared. An internal platform based on Salesforce Agentforce is being rolled out to around 1,000 advisers, helping them answer client questions, prepare recommendations and manage daily work.
This is a more consequential role than a basic chatbot. Advisers sit at the centre of high-value client relationships, and their recommendations depend on current data, regulatory obligations and professional judgment.
The deployment tests whether AI can increase the capacity of that role without weakening the human review on which trust depends.
Agents enter the preparation loop
Earlier banking AI often handled simple support questions or internal productivity tasks. The new platform can analyse client information and assist in forming a response or recommendation in real time.
For an adviser, much of the working day involves gathering records, preparing for meetings and locating relevant information. An agent can reduce that preparation burden and present material in a consistent format.
Bank of America already uses AI at considerable scale. Its Erica virtual assistant is said to perform work equivalent to about 11,000 employees. The bank also reports that 18,000 developers use coding tools associated with roughly 20% productivity improvement.
Advisory work raises different risks. A coding suggestion can be tested before release. A financial recommendation may influence a client’s savings or investment decisions, making context and explanation essential.
Productivity shifts the adviser’s role
Major banks including JPMorgan, Wells Fargo and Goldman Sachs are also testing AI for staff productivity and client-facing work. A common objective is to increase output without expanding headcount at the same pace.
If agents handle research and routine preparation, advisers can spend more time understanding goals, discussing trade-offs and maintaining relationships. The job may become less administrative and more focused on communication and judgment.
That outcome is not automatic. Employees need training to challenge outputs rather than accept them because they are fast and polished. Poor data or incomplete integrations can produce recommendations that look comprehensive while missing critical context.
The source notes that some observers consider current banking AI progress relatively unexciting from a product perspective. That may be precisely what operational adoption looks like: gradual changes to preparation and workflow rather than a dramatic new consumer service.
Human oversight is part of the product
Financial institutions must ensure recommendations meet compliance standards and can be explained if regulators or clients question them. This limits how much autonomy an advisory agent should receive.
The adviser remains responsible for checking relevance, accuracy and suitability. Systems should show the data and assumptions behind a suggestion, record changes and flag uncertainty rather than present one confident answer.
Over-reliance is a distinct risk. As the platform becomes familiar, staff may reduce critical review. Monitoring should therefore examine not only model errors, but how people interact with the tool and whether they detect problems.
Clean, structured data remains a prerequisite. Large banks hold information across products and systems, and connecting it safely can take longer than deploying the model itself. Permissions must also ensure an adviser sees only the client information appropriate to the role.
Bank of America’s initial group of 1,000 advisers is large enough to reveal practical behaviour while still allowing a staged rollout. The bank can compare preparation time, client outcomes, corrections and adoption patterns before expanding.
AI agents are entering banking roles, but not as independent advisers. They are becoming part of the machinery through which professionals prepare and decide.
The strongest result would be almost invisible: better-informed conversations, less time spent searching and a human adviser with more attention available for the client. Achieving that requires treating oversight not as a temporary safeguard, but as a permanent feature of AI-assisted advice.
