Coforge Develops Framework To Guide Enterprise AI Agent Decisions

Coforge has introduced an intent engineering framework for enterprises deploying AI agents, adding a layer focused on defining what autonomous systems should optimise for, how their performance should be measured, and when they should defer to human oversight.

The framework distinguishes between three layers of enterprise AI. 

Prompt engineering influences how a model responds, while context engineering provides the information needed to make decisions. Intent engineering creates the governance structure for those decisions.

Coforge’s approach adds objectives, constraints, and measurable outcomes to determine what an agent should pursue.

“Most enterprises have invested heavily in teaching AI what to say and what it should know,” Vic Gupta, Executive Vice President at Coforge, said in a statement. “The harder challenge is deciding what the agent should optimise for when objectives conflict, risk increases, or exceptions occur.”

The company said the framework includes outcome-based design principles, controls built into the development lifecycle, escalation pathways for high-risk scenarios, and success metrics based on business outcomes rather than activity volume.

The issue of agent boundaries is becoming more relevant as enterprises give AI agents access to business systems and tools. A recent ServiceNow incident involving an AI agent that attempted to change employee salaries highlighted the risks of giving autonomous systems access to sensitive actions.

Coforge said the framework draws on its work on enterprise AI programmes. It is intended to address situations in which business objectives conflict, risk increases, or an agent encounters an exception requiring human intervention.

The company has also been expanding its focus on operating AI agents in production. In September, Coforge expanded its AgenticOps capabilities to address issues including governance, agent drift, security exposure, tool integration, and observability as enterprises move agentic AI systems beyond experimentation.