StartupXO
Language

Language

AI & Automation

AI Agents Just Got Their Own Department at a Korean Brokerage

Published: 2026-08-03

AI AgentsFinancial AIOrg DesignBrokerageHuman-in-the-Loop

What Happened

Shinhan Securities, one of South Korea’s largest brokerages, has created an “AI Agent Department” whose members are not people. According to an August 3 Yonhap report, a lead agent assigns work and three staff agents execute it, covering market research, issue analysis, and drafting meeting materials and proposals. One human AI specialist is attached part-time to make final judgment calls and review output. It is the first standalone department at a Korean financial firm staffed entirely by AI agents. The unit starts with a pilot supporting the wholesale (institutional sales) division, and CEO Lee Sun-hoon says the collaboration will extend to research, wealth management, and investment banking.

This did not come out of nowhere. Shinhan Securities stood up a dedicated AI transformation division in its January reorganization, citing Korea’s new AI Framework Act among the reasons. Its parent, Shinhan Financial Group, is running a “one department, one agent” campaign across subsidiaries, and Shinhan Bank exhibited 18 internally built agents plus 3 from affiliates at its AX Agent Summit on July 21. The group has moved past adopting tools and started writing agents into the org chart.

What This Means for Founders

When an agent lands on the org chart, what the enterprise buys changes. A copilot in a chat window is a personal productivity tool. A registered department of agents needs everything a human team needs: work assignment and progress tracking, output review, performance evaluation, permissions and access control, and an audit trail of who did what, when, and why. Human organizations run on decades of layered HR, groupware, and performance SaaS. The equivalent stack for agent organizations is essentially empty, and in regulated industries like finance that gap is the bottleneck that decides adoption speed.

Just as important, a buyer now exists. Dedicated AI divisions have budgets and sign-off authority. Agent evaluation tools, output verification, and compliance logging finally have a counterpart on the org chart to sell to. The design of Shinhan’s model is worth reading closely too. Agents execute, a human reviews: which means review is the bottleneck. One part-time specialist has to clear the output of four agents, so whatever raises review speed raises the department’s throughput. That is a product category, not a feature.

What You Can Do Now

If you have sold into financial institutions, go after the operating stack of agent departments: evaluation, review, and audit layers first. Demo the management console, not the agent. Show a screen that says what this department did this month and which deliverables got rejected, and the budget holder leans in. Watch the diffusion order in regulated industries as well. A brokerage moved first; when banks and insurers follow, they will demand compliance features you can build into the product now, before the RFPs spell them out.