Wall Street’s AI hiring boom needs people who can manage teams of agents

Even automating an employee’s vacation request can become a maze of exceptions for a bank. That everyday problem helps explain why Wall Street’s AI push is creating demand for people, not just software. Banks need engineers who can coordinate specialized AI agents—systems assigned individual tasks—and decide when a human should step in. References to that skill, known as agent orchestration, have jumped 1,721% this year, according to hiring-data firm Draup’s analysis provided exclusively to CNBC.

AI-related job listings at banks including JPMorgan Chase, Citigroup and Capital One rose 49% this year compared with 2025, reaching 139,819, according to CNBC Economy (direct), citing Draup. The firm draws on public job advertisements and platforms including LinkedIn. The figures measure advertised roles and references to skills, rather than completed hires, but they show where banks are directing their search for talent.

From building models to getting work done

The hiring push reflects a change in what banks want from AI. Earlier recruitment centered on engineers and data scientists who built models or adapted them to company information. Now, banks also want people who can put the technology to work inside individual business units. These workers, often called forward-deployed engineers, combine technical expertise with knowledge of a particular operation, whether a trading desk, an administrative function or human resources.

Their work can involve linking several agents into one process: one examines raw data, another reads a document and a third checks compliance with regulations. Engineers must choose the agents, define their responsibilities, select the technology and establish where human oversight belongs. Draup chief executive Vijay Swaminathan told CNBC that hidden complexity can make even apparently simple processes difficult to automate. Vacation approvals, for example, can involve numerous unusual cases and exemptions.

“This is arguably the hottest skill on Wall Street,” Draup CEO Vijay Swaminathan told CNBC, referring to agent orchestration.

Demand is also rising for the tools behind these workflows. References to LangGraph, a framework for building processes with multiple steps, increased 679%. Mentions of LlamaIndex, which connects AI applications to data, rose 291%. Retrieval-augmented generation, or RAG—a technique that supplies AI models with information from company databases—recorded a 259% increase. Swaminathan said employers also increasingly value problem-solving, creativity and the confidence to question how a process really works.

Controls and talent become the bottlenecks

Banks are recruiting for safeguards alongside automation. Job-post references to responsible AI rose 657%, while mentions of AI governance and risk management increased 394% and 359%, respectively. Governance-related skills accounted for more than 16,000 references in Draup’s data, compared with roughly 8,400 associated with training, deploying and operating models. Swaminathan said cybersecurity teams are focused on ensuring that outside tools and connections to external models do not introduce vulnerabilities across their systems.

The specialized work commands a premium, but pay alone has not solved the staffing challenge. Generative AI managers—overseeing technology that creates content—receive median base salaries of about $190,000, according to Draup. Roles involving generative AI and agents generally pay more than technology jobs elsewhere in finance. Even so, Swaminathan said these positions remain difficult to fill, pushing major banks to rely heavily on programs that retrain existing developers and business specialists.

The next phase to watch is how that internal training translates into deployment and changes to employees’ work. JPMorgan CEO Jamie Dimon has discussed extensive redeployment plans as AI handles more tasks. For banks, the challenge is not simply finding people who understand the technology. It is finding—and training—people who understand the business well enough to make automation useful, while keeping the necessary human checks in place.