Enterprises are shifting their focus from merely utilizing AI agents to ensuring these agents can effectively collaborate with one another. This shift highlights the importance of orchestration across multi-agent systems, which is becoming a vital differentiator in the business landscape. 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 arise, akin to people speaking different languages, which ultimately diminishes the quality of actions and can lead to serious issues such as data leakage.
Enhancing Coordination Among AI Agents
Traditionally, orchestration focused primarily on data, but this is evolving toward managing actions. Solutions that act like conductors are increasingly integrating AI agents, Robotic Process Automation (RPA), and data repositories. Sanders likened this evolution to the development of answer engine optimization, which has transitioned from monitoring to creating tailored content and code. These orchestration platforms are designed to coordinate various agentic solutions, thereby enhancing the consistency of outcomes.
Among the early providers of such tools are Salesforce MuleSoft, UiPath Maestro, and IBM Watsonx Orchestrate. Their “phase one” software-based observability dashboards assist IT leaders in monitoring all agent actions across an organization, which has become crucial for effective risk management.
Addressing Risks in Coordination
While coordination is essential, its value is limited without robust risk management features. Future orchestration platforms are expected to evolve into technical risk management tools that enhance quality control. These systems could include functionalities for agent assessments, policy recommendations, and proactive reliability scoring. Many enterprise leaders are cautious about depending solely on vendors to mitigate risks. According to Sanders, “Many IT decision-makers do not trust a vendor’s statements about the reliability of their agents.”
To address these concerns, third-party tools are emerging to automate tedious processes, such as managing guardrail approvals and escalation tickets. For instance, in banking, the loan approval process typically requires 17 steps, and AI agents often disrupt human workflows by seeking approval at established guardrails. Third-party orchestration platforms can streamline these interactions, potentially eliminating the need for continuous human oversight and allowing organizations to realize significant efficiency gains.
“True velocity gains” could be achieved, Sanders noted, where outcomes are measured not simply in percentages but in multiples, such as 3X versus 30% improvement in efficiency.
The Shift Towards Human-On-The-Loop Design
The role of human evaluators in the agentic era is also evolving. Rather than being merely “human-in-the-loop,” these individuals will transition to “human-on-the-loop,” designing agents to automate workflows. With continuous advancements in no-code solutions, nearly anyone can create an agent using natural language, as noted by Sanders. This democratization of agentic AI requires individuals to articulate goals, provide context, and foresee potential issues, mirroring the skills of effective people managers.
To capitalize on this trend, organizations should initiate “expeditious programs” that integrate agents across workflows, particularly in areas with repetitive tasks that create bottlenecks. In the early stages, a strong human-in-the-loop element will be necessary to ensure quality and manage change effectively. Sanders emphasizes that serving as evaluators will deepen the understanding of these systems, enabling teams to operate upstream in agentic workflows rather than downstream.
IT leaders are encouraged to take stock of their entire automation stack, which includes rules-based automation, RPA, and agentic automation. Understanding all components will be essential to maximize the potential of emerging orchestration platforms. Sanders warns that failing to do so could lead to dis-synergies, where legacy systems clash with cutting-edge technologies, particularly in customer-facing scenarios. “You can’t orchestrate what you can’t see clearly,” he concluded.