IBM sells real-time dashboards to supervise AI agents
Consulting shifts from expert labour to monitored workflows and audit trails, accountability migrates to the human sign-off layer
Images
IBM is pushing a new management interface for the age of “digital employees”: a real-time dashboard that lets human staff monitor chains of AI agents as they perform consulting work.
Business Insider reports that IBM has been using such a dashboard internally and released it to clients earlier this year. The company says AI agents have sped up security investigations, cutting work that used to take around 45 minutes down to a few minutes. IBM Consulting’s revenue reached $21 billion in 2025, the outlet notes, with demand for AI solutions helping drive growth.
The immediate appeal is cost and throughput. If an agent can triage incidents, draft documentation, or correlate logs faster than a junior analyst, a consulting firm can sell “more output” without scaling headcount at the same rate. A dashboard also helps sell reassurance: the client is not just buying an opaque model, but a control room—status indicators, task queues, and audit trails—that resembles familiar enterprise governance.
But the dashboard solves a business problem as much as a technical one. When work is done by agent chains—model A calls tool B, which triggers workflow C—errors become harder to attribute. The vendor can point to the customer’s configuration, the customer can point to the vendor’s agent, and the human “supervisor” becomes the person whose name ends up on the incident report. The monitoring layer formalises that handoff: it turns oversight into a job role, and it makes “human in the loop” a compliance feature rather than a guarantee of understanding.
That shift matters because consulting is not only advice; it is liability management. Clients buy a large firm partly to outsource risk and to obtain documentation that will satisfy regulators, auditors, and boards. If AI agents produce the work product, the value of the human consultant increasingly becomes sign-off, traceability, and defensible process—checking prompts, validating outputs, and preserving logs. The work moves from producing analysis to producing an evidentiary trail.
The second-order effect is a new labour split inside white-collar firms. A small number of people design workflows and choose tools; a larger group watches dashboards, handles edge cases, and absorbs blame when automation fails. The promise of efficiency remains, but the organisation still needs humans—just in roles that look less like expertise and more like operational insurance.
IBM’s pitch is that clients can watch AI agents work in real time, even as the human whose job is to watch them becomes the default owner of the consequences.