Climate researchers are increasingly turning to artificial intelligence to address the challenges of climate change. However, a growing concern is that the current approach often starts with what AI can do today and then seeks ways to apply it to climate problems. This backward logic risks misaligning AI capabilities with the real needs of climate management. Instead, we must reverse the workflow: begin by asking what climate change mitigation and adaptation truly require from AI, and then design systems accordingly. The urgency of this shift is clear. Climate change demands solutions that are not only technically robust but also socially equitable, scalable, and aligned with long-term sustainability goals. Yet many AI applications in climate science focus narrowly on improving forecasts or optimizing energy systems without considering broader systemic challenges—such as policy implementation, behavioral change, or justice in climate adaptation. A more purpose-driven approach would prioritize AI systems that support decision-making under uncertainty, integrate diverse data sources (including Indigenous knowledge and local observations), and enable transparent, accountable governance. For example, AI could help identify vulnerable communities, simulate the impacts of different policy interventions, or support real-time monitoring of carbon sequestration efforts. But to do so effectively, AI must be co-designed with climate experts, policymakers, and affected communities. Moreover, the current emphasis on model performance metrics—such as accuracy or speed—can overlook critical trade-offs. An AI system that delivers high precision but lacks interpretability or fairness may undermine trust, especially in high-stakes climate decisions. We need to build models that are not just powerful, but also explainable, auditable, and resilient to bias. The path forward requires collaboration across disciplines. Computer scientists must work closely with climate scientists, economists, and social scientists to define meaningful objectives. Funding agencies and research institutions should prioritize projects that are driven by climate needs rather than AI capabilities. And regulatory frameworks must evolve to ensure that AI is used ethically and effectively in climate action. In short, the goal should not be to use AI for climate change, but to build AI that is truly fit for the purpose of climate management. By rethinking how we develop and deploy AI, we can create tools that are not only smarter, but also more just, inclusive, and effective in safeguarding the planet.