Instead of asking how AI agents can do the work for them, a key question in enterprises now is: Are agents playing well together?
This makes orchestration on multi-agent systems and platforms a key concern – and a key differentiator.
“Agent-to-agent communication is really emerging as a big deal,” Tim Sanders, chief innovation officer at G2, told VentureBeat. “Because if you don’t organize it, you get misunderstandings, like people speaking foreign languages to each other. Those misunderstandings reduce the quality of work and increase the risk of hallucinations, which can lead to security incidents or data leakage.”
Allowing agents to talk and coordinate
Orchestration up to this point has largely been around data, but this is increasingly turning into action. “Conductor-like solutions” are increasingly bringing together agents, robotic process automation (RPA), and data repositories. Sanders compared the progress to Answer Engine optimization, which initially started with monitoring and now creates custom content and code.
“Orchestration platforms coordinate different agentic solutions to increase consistency of outcomes,” he said.
Initial providers include Salesforce MuleSoft, UiPath Maestro, and IBMwattsonx Orchestrate. These “step one” software-based overview dashboards help IT leaders see all agentive activities across an enterprise.
Key elements of risk management
But coordination can only add so much value; These platforms will turn into technical risk management tools that will provide better quality control. For example, this could include agent evaluation, policy recommendation, and proactive scoring (e.g., how reliable agents are when they call enterprise tools, or how often and when they hallucinate).
Enterprise leaders have become wary of relying on vendors to mitigate risks and errors; Many IT decision-makers, he said, do not, in fact, trust vendor statements about the reliability of their agents.
Third-party tools are beginning to bridge the gap and automate the arduous railing processes and escalation tickets. Teams are already experiencing “ticket fatigue” in semi-automated systems, where agents hit guardrails and require human permission to proceed.
For example: The loan approval process at a bank requires 17 steps, and when it runs into established guardrails an agent keeps interrupting the human workflow with approval requests.
Third-party orchestration platforms can manage these tickets and even override the need for approval altogether. They can ultimately eliminate the need for constant human-in-the-loop oversight so organizations can experience “true velocity gains” that are measured not in percentages but in multiples (i.e., 3X vs. 30%).
“Where it goes is remote management of the entire agentic process for organizations,” Sanders said.
‘Human in the Loop’ vs. ‘Human in the Loop’
According to Sanders, in another important development in the agentic era, human evaluators will become designers, moving from human-in-the-loop to human-on-the-loop. Namely: they will start designing agents to automate workflows.
Agent builder platforms continue to innovate their no-code solutions, Sanders said, meaning almost anyone can now stand up an agent using natural language. “This will democratize agentic AI, and the super skill will be the ability to articulate a goal, provide context, and visualize pitfalls, which is the equivalent of a good people manager today.”
What should enterprise leaders do now
Agent-first automation stacks “dramatically outperform” hybrid automation stacks in almost every attribute, he said: satisfaction, quality of tasks, security, cost savings.
Organizations should launch “accelerated programs” to incorporate agents into the workflow, especially with highly repetitive work that creates bottlenecks. Perhaps first, there will be a strong human-in-the-loop element to ensure quality and promote change management.
“Working as an evaluator will strengthen our understanding of how these systems work,” Sanders said, “and will ultimately enable us all to work upstream in agentic workflows rather than downstream.”
IT leaders should take inventory of all the different elements of their automation stack today. Whether these elements are rules-based automation, RPA, or agentic automation, they need to learn everything going on in the organization to make optimal use of emerging orchestration platforms.
“If they don’t do that, it can really lead to disharmony in organizations where old-school technology and cutting-edge technology collide at the point of delivery, many times customer-facing,” Sanders said. “You can’t organize what you can’t see clearly.”
<a href