It's like having a whole team working with you, and each of these agents is going to be a specialist in a particular area of your work. It’s just too much for one single person to handle.
The age of doing is giving way to the age of directing. No longer just assistants, AI agents are becoming autonomous teammates, taking on entire goals and reshaping what it means to be a knowledge worker.
Jorge Contreras, Applications Manager at marine advisory and brokerage firm McQuilling Partners, Inc., is treating AI agents less like tools and more like colleagues joining the team.
Agents on board: "It's like having a whole team working with you," says Contreras, "and each of these agents is going to be a specialist in a particular area of your work." The shift is already in motion, with professionals moving from doing the work to directing a digital team. In industries like his, where brokers juggle overwhelming volumes of client information, AI agents offer much-needed scale. “It’s just too much for one single person to handle.”
Speaking the same language: Contreras points to the emergence of standards like the MCP as the abstraction layer that allows different tools to communicate and track progress towards a unified goal. "The MCP acts as the Rosetta Stone for our entire toolchain," Contreras says. It provides a crucial layer so that individual tools can evolve or be swapped out without breaking the entire workflow. As long as a tool can speak MCP, it can communicate its status and results, allowing teams to track progress in real-time.
Trust but verify: But managing this new team of AI agents comes with unique challenges that go beyond typical software bugs. Contreras highlights a more subtle risk: an agent learning the wrong thing and embedding it into its memory, leading to a cascade of flawed outputs. “If an agent is storing an invalid assumption in its memory, it's going to produce some invalid suggestions and recommendations,” he warns. "It's a twofold story. Culturally, it's about educating people. Technologically, it's about building safeguards to keep the system under our control."
Skip the prompt: Contreras predicts the interaction model will flip from reactive to proactive. “Imagine you log in in the morning and the system already has a summary for you,” he says. “The tools and information you usually need to do your job are presented ahead of time in a dashboard, so you don't even have to ask the agent for it. It’s done for you in advance.”
