“In dynamic enterprise environments with frequent UI changes, these agents risk brittleness unless paired with augmented data management, adaptive retraining, and fallback mechanisms; therefore, at this stage, they are more suited for controlled workflows than mission-critical automation,” Dai said.
Performance is only one part of the equation. Enterprises will also need stronger controls before allowing such agents to run unsupervised on internal systems.
“These agents are convenient, but a rogue action could cause damage,” Jain added. “You need strong governance frameworks in place before deploying them at scale.”
Sheel said firms should define clear human-oversight points, such as when “Critical Points” arise, maintain audit trails for every action the agent takes, enforce role-based access controls, and monitor performance and errors continuously. “They should also include a remediation strategy for when the agent makes mistakes or behaves undesirably, and ensure data governance, privacy, and compliance policies are built into the agent’s workflows,” Sheel added.
