The integration of AI agents into business operations is evolving, with a growing emphasis on how these agents communicate and collaborate. Instead of merely evaluating their individual effectiveness, organizations are now focusing on whether these agents can work together efficiently. This shift highlights the importance of orchestration in multi-agent systems, which has become a critical factor in successful enterprise operations.
Tim Sanders, Chief Innovation Officer at G2, recently emphasized this point in an interview with VentureBeat. He noted, “Agent-to-agent communications is emerging as a really big deal. If you don’t orchestrate it, you get misunderstandings, like people speaking foreign languages to each other.” These miscommunications can lead to reduced action quality and increase the risks of data leaks or security incidents.
Transitioning from Data to Action
Historically, orchestration in AI systems has centered around data management. However, this focus is rapidly shifting towards action coordination. Emerging solutions, described by Sanders as “conductor-like,” are beginning to integrate multiple agents, Robotic Process Automation (RPA), and data repositories.
This transition mirrors the evolution of answer engine optimization, which began with monitoring before advancing to the creation of tailored content and code. Orchestration platforms are now designed to synchronize various agentic solutions to enhance the consistency of outcomes. Key players in this space include Salesforce MuleSoft, UiPath Maestro, and IBM Watsonx Orchestrate. These initial software-based observability tools allow IT leaders to track all agent activities within their organizations.
Addressing Risk Management Concerns
While coordination is essential, its value is limited without addressing the underlying risks associated with agent operations. Future iterations of orchestration platforms are likely to evolve into sophisticated technical risk management tools, providing enhanced quality control. This could include functionalities such as agent assessments and proactive scoring to evaluate reliability when agents interact with enterprise tools.
In a climate of skepticism regarding vendor assurances, many IT decision-makers are cautious about the reliability of their agents. As Sanders pointed out, third-party tools are beginning to fill this gap, automating tedious guardrail processes and reducing the “ticket exhaustion” experienced in semi-automated systems. A common challenge arises in processes like bank loan approvals, which may involve up to 17 steps. Agents often disrupt human workflows with unnecessary approval requests when encountering established guardrails. Third-party orchestration platforms can mitigate these interruptions by managing approval tickets, potentially eliminating the need for constant human oversight.
Sanders predicts that this evolution will lead to increased operational efficiency, with organizations experiencing significant velocity gains. He stated, “Where it goes from there is remote management of the entire agentic process for organizations.”
The concept of “human-on-the-loop” is emerging as a critical development in agent management. Instead of merely overseeing the process, human evaluators are expected to take on more of a design role, creating agents that can automate workflows. Innovations in no-code solutions are enabling more individuals to create agents using natural language, democratizing access to agentic AI. This shift emphasizes the importance of clearly expressing goals, providing context, and anticipating potential pitfalls.
Organizations are encouraged to initiate “expeditious programs” to integrate agents into their workflows, particularly in areas where repetitive tasks create bottlenecks. Initially, a strong “human-in-the-loop” presence will be necessary to ensure quality and facilitate change management. Sanders noted that serving as evaluators can enhance understanding of these systems, ultimately allowing teams to operate upstream in agent workflows.
Finally, IT leaders are advised to inventory the various elements of their automation stacks, which may include rules-based automation, RPA, or agentic automation. Understanding the entire automation landscape is vital for optimal use of emerging orchestration platforms. “If they don’t, there could actually be dis-synergies across organizations where old school technology and cutting-edge technology clash at the point of delivery,” Sanders warned. “You can’t orchestrate what you can’t see clearly.”
As organizations continue to grapple with the complexities of AI agents, effective orchestration will undoubtedly be a defining factor in their success.
