Rather than asking how AI agents can work for them, a key question in businesses now is: do agents play well together?
This makes orchestration across multi-agent systems and platforms a critical concern – and a key differentiator.
“Agent-to-agent communications is becoming a real problem,” Tim Sanders, G2’s chief innovation officer, told VentureBeat. “Because if we don’t orchestrate it, we get misunderstandings, as if people were speaking foreign languages to each other. These misunderstandings reduce the quality of actions and give rise to the specter of hallucinations, which could be security incidents or data leaks.”
Allow agents to talk and coordinate
Until now, orchestration has relied largely on data, but it is quickly transforming into action. “Driver-like solutions” increasingly bring together agents, robotic process automation (RPA), and data repositories. Sanders compared this progression to that of response engine optimization, which initially started with monitoring and now creates custom content and code.
“Orchestration platforms coordinate a variety of different agent solutions to increase consistency of results,” he said.
Early vendors include Salesforce MuleSoft, UiPath Maestro and IBM Watsonx Orchestrate. These “phase one” software-based observability dashboards help IT managers visualize all agent actions within an enterprise.
The critical element of risk management
But coordination can only add limited value; these platforms will transform into technical risk management tools offering better quality control. This can include, for example, agent evaluations, policy recommendations, and proactive scores (e.g. how reliable agents are when engaging company tools, or how often and when they hallucinate).
Business leaders are now reluctant to rely on suppliers to minimize risks and errors; In fact, many IT decision-makers don’t trust vendors’ claims about the reliability of their agents, he said.
Third-party tools are beginning to bridge the gap and automate tedious gatekeeping and escalation ticket processes. Teams already experience “ticket burnout” in semi-automated systems, where officers hit guardrails and need human authorization to continue.
As an example: The loan process at a bank requires 17 steps for approval, and an agent continues to interrupt human workflows with approval requests when running into established guardrails.
Third-party orchestration platforms can manage these tickets and not, or even challenge the need for approval altogether. They can ultimately eliminate the need for persistent human monitoring so that organizations can benefit from “real speed gains” measured not in percentages but in multiples (i.e. 3x versus 30%).
“From there, there’s remote management of the entire agent process for organizations,” Sanders said.
“human in the loop” versus “human in the loop”
In another critical evolution of the agentic era, human evaluators will become designers, moving from human-in-the-loop to human-in-the-loop, according to Sanders. In other words: they will start designing agents to automate workflows.
Agent creation platforms continue to innovate with their no-code solutions, Sanders said, meaning almost anyone can now represent an agent using natural language. “This will democratize agentic AI, and the super skill will be the ability to express purpose, provide context and consider pitfalls, which is very similar to that of a good HR manager today.”
What Business Leaders Should Do Now
Agent-centric automation stacks “significantly outperform” hybrid automation stacks in almost every attribute, he noted: satisfaction, action quality, security, cost savings.
Organizations should launch “rapid programs” to integrate agents into all workflows, especially for highly repetitive tasks that pose bottlenecks. It is likely that initially there will be a strong human element in the loop to ensure quality and promote change management.
“Serving as an evaluator will increase understanding of how these systems work,” Sanders said, “and ultimately allow all of us to operate upstream in agent workflows rather than downstream.”
IT leaders should take inventory of all the different elements of their automation stack today. Whether it’s rules-based automation, RPA, or agentic automation, they need to learn everything that’s happening in the organization to optimally use emerging orchestration platforms.
“If they don’t do this, there could be dissynergies between organizations where old technology and new technology clash at the point of delivery, often facing the customer,” Sanders said. “You can’t orchestrate what you can’t see clearly.”
