Five questions to answer before AI agents enter production

A successful AI pilot does not show that an organization is ready for production. ManageEngine CEO Rajesh Ganesan says IT leaders should answer five questions before giving an AI agent access to live systems.

AI agents can already detect anomalies in large datasets, document incidents and help with recurring support work. Ganesan says they should not yet change policies, deploy software or make other decisions where an error could have major consequences. The safest role is an observer working alongside a person, with a human retaining responsibility.

Ganesan describes a gradual path from passive monitoring to active analysis and then limited actions. ManageEngine, part of Zoho Corporation, has begun rolling out AI agents across products for IT operations, ITSM, endpoint management and cybersecurity. The company uses AI for code generation, but people review every line it creates.

Metro Service A/S, which operates and maintains Copenhagen’s metro, uses ManageEngine products including ServiceDesk Plus. Its AI summarizes long support conversations and helps document solutions, while much of its automation still relies on predefined rules. Piotr Wojciech Mankiewicz says the organization is “very far” from letting AI make operational decisions because errors and data-hygiene problems remain.

Ganesan says a pilot only shows whether an organization is ready to begin. Before production access, data, permissions, agent identities, actions and responsibility for mistakes must be clear.