How agentic AI drives net-zero goals through innovation and governance

A new study has found that agentic artificial intelligence helps companies cut emissions and reach climate goals, but the size of the gain depends on how the firm innovates, how quickly it can reconfigure operations, and how it builds ethics into everyday workflows. The authors also show that outside conditions, from policy to market complexity, can weaken these benefits if leaders do not plan for them.

The research, titled “Innovation Dynamics and Ethical Considerations of Agentic Artificial Intelligence in the Transition to a Net-Zero Carbon Economy” and published in Sustainability, reports survey data and model tests that link agentic AI to better climate results through specific pathways inside the firm.

The team surveyed 340 organisations operating across energy, manufacturing, agriculture, transport, and related sectors. Firms were included only if they had at least 50 employees, real experience with agentic AI, and active climate initiatives, and only if a senior technical leader could answer the survey. The final sample was drawn from a screened pool of more than 500 contacts and reflects companies already deploying agentic AI for climate tasks.

The authors’ model tracks six linked parts, from the firm’s outside environmental context, to dynamic capabilities, agentic AI, innovation dynamics, ethical considerations, and climate performance. This framework sets the logic for the tests that follow.

How much does agentic AI move the needle on climate results?

The study reports a clear positive link between agentic AI capabilities and climate mitigation outcomes. In the structural tests, agentic AI shows a strong direct association with climate performance, backed by additional regressions and robustness checks. The authors validate their measures with confirmatory factor analysis, and they address common method bias through design steps and diagnostic tests.

Agentic AI can help firms detect climate risks, seize low-carbon opportunities, and reconfigure operations at speed. On its own, though, the technology does not guarantee impact. The largest gains appear when AI sits inside a firm that already senses change well, decides quickly, and can shift resources without friction.

What inside the firm turns AI potential into real climate gains?

The data show that innovation dynamics, dynamic capabilities, and ethical considerations act as bridges between AI and climate performance. All three produce significant indirect effects, which means they help carry the effect of AI to the final outcome. The bootstrapped confidence intervals for these mediators exclude zero, supporting the main hypotheses on mediation.

In practical terms, the results reward firms that invest in agile teams, rapid testing, cross-functional work, and continuous capability renewal. Where these features are strong, AI-driven sensing and action feed directly into measurable climate benefits. The authors’ multi-model regressions show rising explanatory power as mediators enter the model set, which aligns with the narrative that process and capability are the channels that matter.

The paper also stresses measurement quality. Factor loadings exceed accepted cutoffs, average variance extracted and composite reliability meet standards, and global fit indices clear common thresholds, which supports confidence in the constructs used for agentic AI, innovation dynamics, dynamic capabilities, ethics, and climate outcomes.

When do ethics and outside conditions help, and when do they slow progress?

Ethics play a dual role. The study finds that ethical considerations strengthen trust, transparency, and accountability, and they are part of the indirect pathway from AI to climate results. At the same time, strict ethics processes can slow approvals and delay climate projects. To fix this, the authors propose three steps firms can adopt now. First, adaptive ethical protocols that scale oversight to the level of risk. Second, pre-approved ethical templates that can cut approval time by about 60 percent. Third, stakeholder co-design so teams build agreement early in the project.

The external setting also matters. The analysis shows that the environmental context weakens the mediation paths through innovation dynamics and ethics. In places with fragmented rules, fragile infrastructure, or high uncertainty, the chain from AI to results is less efficient. The moderated-mediation indices are negative and significant for the innovation and ethics channels, which signals a real dampening effect from context.

For leaders, the policy take-away is direct. Build governance that keeps ethics strong while avoiding needless delays, and plan for context risk. The paper suggests robust stakeholder engagement, clear internal rules, and early capability building, because once AI-based dynamic capabilities mature, climate performance becomes more resilient to outside turbulence.

Additional details that matter to decision-makers

• Sample quality: The authors screened out small firms, companies without clear climate targets, and responses without access to senior technical leaders. The design reduces noise and keeps the focus on firms actually using agentic AI for climate tasks.

• Bias checks: The study reports tests for non-response bias and common method bias. Results suggest these are not a major concern in the dataset.

• Limits: Findings best apply to medium and large firms with existing AI capacity, often in faster-growing markets. The cross-section design limits causal claims and may not generalise to highly regulated industries or very small companies.

To sum up, agentic AI is not a switch you flip. It is a lever that works when the firm has the right muscles. The study shows that the biggest climate gains come when companies combine agentic AI with fast and disciplined innovation, strong dynamic capabilities, and ethics that protect people while keeping projects moving. Leaders who tune all three, and who plan for tough external settings, are more likely to turn AI into real progress on net-zero goals.