Organizations are increasingly focusing on whether artificial intelligence (AI) agents can effectively collaborate rather than solely on how these agents can perform tasks. The orchestration of multi-agent systems has emerged as a significant concern for enterprises, influencing how AI technologies are integrated and managed. According to Tim Sanders, Chief Innovation Officer at G2, “Agent-to-agent communications is emerging as a really big deal.” Without proper orchestration, misunderstandings can occur, akin to people speaking different languages, which can lead to decreased action quality and potential security risks.
Enhancing Coordination Among AI Agents
Traditionally, orchestration has centered on data management. However, the focus is shifting toward actionable strategies. Solutions that function like conductors are gaining traction, integrating agents, robotic process automation (RPA), and data repositories. Sanders compares this evolution to the development of answer engine optimization, which has progressed from basic monitoring to the creation of tailored content and code. He emphasizes that orchestration platforms are designed to coordinate various agentic solutions, ultimately enhancing the consistency of outcomes. Early providers in this space include Salesforce MuleSoft, UiPath Maestro, and IBM Watsonx Orchestrate. These platforms serve as observability dashboards, allowing IT leaders to monitor agent activities across their organizations.
Despite these advancements, coordination alone may not suffice. Sanders notes that these orchestration platforms will evolve into technical risk management tools, enhancing quality control by offering features such as agent assessments and policy recommendations. Many enterprises have grown skeptical of their vendors’ claims regarding reliability, prompting a demand for third-party tools that can automate tedious processes and escalate issues when necessary.
For instance, in a banking context, a loan approval process may comprise 17 steps, with agents frequently interrupting human workflows for approvals when encountering established guardrails. Third-party orchestration platforms can help manage these interruptions, assess the need for approvals, and ultimately reduce the reliance on human oversight. This shift could lead to significant efficiency gains, allowing organizations to operate at multiple times their current velocity.
Shifting Roles in AI Management
A significant transformation is on the horizon as human evaluators transition from merely overseeing processes to actively designing agents. This evolution moves from a “human-in-the-loop” approach to a “human-on-the-loop” model, where individuals will create agents capable of automating workflows. Sanders points out that advancements in no-code solutions will enable almost anyone to develop an agent using natural language, democratizing access to agentic AI. The new skill set will focus on articulating goals, providing context, and identifying potential pitfalls, akin to effective people management.
For enterprise leaders, the immediate priority should be to implement programs that integrate agents into workflows, particularly in areas with repetitive tasks that create bottlenecks. Initially, a human-in-the-loop component will be vital to ensure quality and facilitate change management. Sanders asserts that actively participating as evaluators will deepen understanding of how these systems function, ultimately allowing organizations to work more efficiently in agentic workflows.
IT leaders are encouraged to take stock of their current automation landscape, encompassing rules-based automation, RPA, and agentic automation. Understanding the existing elements is crucial for maximizing the potential of emerging orchestration platforms. Failure to do so could result in dis-synergies, where outdated technology clashes with cutting-edge solutions, particularly in customer-facing scenarios. As Sanders aptly states, “You can’t orchestrate what you can’t see clearly.”
