Soils are critical for storing carbon, sustaining ecosystems, and underpinning global food and water systems. However, managing these complex resources becomes increasingly difficult as climate pressures and land use intensify. A new paper in Frontiers in Science argues that artificial intelligence can help scientists adapt soil systems to a changing climate by making sense of this complexity. Currently, soil science relies on machine learning for tasks like digital soil mapping and spectroscopy. The authors suggest advancing these tools to create digital soil twins, enhanced by sensor data, to monitor soil microbiomes and test climate adaptation strategies in computer models before field implementation. Senior author Professor Alex McBratney of the University of Sydney stated that AI, in partnership with experts, could better match the ever-changing nature of soil ecosystems. Unlike isolated machine learning tools, these advanced systems can mimic scientific collaboration by combining reasoning, planning, and interdisciplinary insights. Lead author Professor Budiman Minasny added that understanding soils can support sustainable agriculture and climate adaptation by helping land managers detect issues like nutrient loss and erosion earlier. To demonstrate this potential, the research team tasked a multi-agent AI system with reviewing scientific literature to generate hypotheses on how soils store carbon. The system successfully produced five distinct hypotheses regarding climate influence, saturation thresholds, and management strategies. These were evaluated through expert opinion and simulated peer review, showing outputs that strongly align with expert research. The study assessed the system's ability to perform perceptual processing, strategic planning, and scientific reasoning, highlighting the promise of multi-agent AI for global soil stewardship. Despite these advances, the paper notes significant challenges, including data quality, model transparency, trust, computational costs, and ethical concerns. Co-author Dr. Mercedes Román Dobarco from NEIKER in Spain emphasized that while AI can emulate some aspects of expert reasoning, it cannot replace the contextual judgment, creativity, and critical interpretation that scientists provide. AI agents also face risks related to dataset bias and interpretability without human oversight. Consequently, the authors advise treating AI as an augmentative tool that enhances rather than replaces human scientific work. The paper underscores AI's ability to accelerate both fast and slow science. By automating time-intensive preparatory tasks like literature reviews, AI can free researchers to focus on foundational understanding and fieldwork while maintaining rigor. Professor McBratney concluded that to fully benefit from AI-enhanced soil science, the scientific community must embrace interdisciplinary collaboration, ensure equitable access to tools, and address ethical challenges. Bridging digital innovation with real-world application, supported by non-negotiable human oversight, could unlock new levels of soil security, provided that human knowledge keeps pace with technological advancement.