The effectiveness of artificial intelligence (AI) agents in enterprise settings hinges on their ability to communicate and collaborate efficiently. As businesses increasingly implement multi-agent systems, the focus has shifted from simply deploying AI agents to ensuring they can work together seamlessly. This orchestration is becoming a critical concern, with significant implications for operational performance.
According to Tim Sanders, Chief Innovation Officer at G2, “Agent-to-agent communications is emerging as a really big deal.” He emphasized that without proper orchestration, misunderstandings can occur, akin to individuals speaking different languages. Such miscommunications can diminish the quality of outcomes and may even lead to hallucinations, which pose risks such as data leakage or security breaches.
Shifting Focus from Data to Action
Historically, orchestration has revolved around data management. However, the current trend is moving towards action-oriented solutions. “Conductor-like solutions” are increasingly integrating agents, robotic process automation (RPA), and data repositories. Sanders likened this evolution to the development of answer engine optimization, which has transitioned from mere monitoring to the creation of tailored content and code.
Orchestration platforms are now being developed to coordinate various agentic solutions, enhancing the consistency of outcomes. Early adopters of such technology include major players like Salesforce MuleSoft, UiPath Maestro, and IBM Watsonx Orchestrate. These “phase one” observability dashboards assist IT leaders in tracking all agentic actions within an organization, providing a comprehensive view of their operations.
Enhancing Risk Management through Orchestration
While coordination adds significant value, these platforms are also evolving into technical risk management tools aimed at improving quality control. Features may include agent assessments, policy recommendations, and proactive scoring that evaluate the reliability of agents when accessing enterprise tools. Many IT decision-makers have expressed skepticism about vendors’ claims regarding the dependability of their agents, prompting a search for third-party solutions that can streamline tedious processes such as escalation tickets.
An example illustrating this challenge is the loan approval process at a bank, which requires 17 steps. An AI agent may frequently interrupt human workflows by requesting approvals when it encounters established guardrails. Third-party orchestration platforms can manage these requests, potentially eliminating the need for constant human oversight, allowing organizations to achieve “true velocity gains.” Sanders notes that these gains could be measured in multiples, such as 3X or even 30%.
Looking ahead, Sanders predicts that human roles will evolve from merely supervising AI (human-in-the-loop) to actively designing these agents (human-on-the-loop). This shift means that individuals will focus on creating agents capable of automating workflows. The continued advancement of no-code agent builder platforms suggests that more people will have the ability to develop agents using natural language, democratizing access to agentic AI.
Strategies for Enterprise Leaders
To maximize the benefits of agent-first automation stacks, which outperform hybrid stacks in various metrics including satisfaction and cost savings, organizations should initiate “expeditious programs” to integrate agents into their workflows, particularly in areas with repetitive tasks that create bottlenecks. In the initial stages, a strong human-in-the-loop component will be essential to ensure quality and facilitate change management.
Sanders advises that IT leaders take a comprehensive inventory of their automation stack, which may include rules-based automation, RPA, and agentic automation. Understanding the full scope of their automation capabilities is vital for effectively utilizing emerging orchestration platforms. Failure to do so could lead to dis-synergies, where outdated and cutting-edge technologies clash, particularly at customer touchpoints. “You can’t orchestrate what you can’t see clearly,” Sanders concludes, highlighting the importance of visibility in the orchestration process.
