Agentic AI for Sustainability Decision-Making in Enterprises

Abstract

Enterprise sustainability decisions require companies to bring together different types of information from the environment, society, governance, finances, and operations. They also need to balance different goals and maintain clear human responsibility. Agentic AI can help with planning, using tools, managing multiple agents, and adapting workflows. However, there is not much guidance on how to properly control these AI features in the context of enterprise sustainability. This study introduces the Enterprise Agentic Sustainability Decision Framework (EASDF) by looking at recent research and analyzing three interviews with experts. Four common themes came up: high-quality data and clear trade-offs between choices; specialized coordination and reliable handoffs between agents; keeping humans accountable and managing decisions in real-time; and ensuring value is achieved and scaled properly. The experts emphasized the need for good data, clear ownership, tracking where data comes from, controlling access, making sure AI models fit the tasks they are used for, getting human approval, keeping records, and measuring the cost-effectiveness of outcomes. The refined EASDF includes eight connected areas covering data foundations, how decisions are framed, trade-offs between stakeholders, managing agents, using tools within limits, who has the right to make decisions, how governance is controlled, and how to provide feedback on results. The framework shows that agency should be gradual and only used when there are enough evidence, control, reversibility, accountability, and shown value. The findings offer a starting point for creating auditable workflows that turn evidence into action, but they do not prove that these approaches are effective, statistically reliable, or fully validated.