What AI Can Already Do for Coherence

Beyond the alignment of the models themselves, there is a parallel question about what intelligence systems are being asked to do. This is where the book’s framing—that intelligence is substrate-sensitive, that the same technology can be aimed at extraction or regeneration depending on the conditions of its deployment—becomes operationally testable.

Consider regenerative agriculture, which has emerged over the past several years as one of the clearest cases where AI is being deployed against the extractive logic that dominates its industry. The McKinsey Global Farmer Insights survey from 2024 reported that 68% of surveyed farmers had adopted crop rotations, 56% had implemented reduced or no-till practices, and 40% were using variable-rate spraying or fertilization. These are not romantic gestures. They are working agricultural practices, supported increasingly by AI systems that can read soil composition, monitor biodiversity indicators, model nutrient cycling, and recommend interventions calibrated to local ecological conditions rather than to industrial uniformity. Digital twin technologies are being used to simulate field conditions before interventions are deployed. Drone-based AI is being used to track populations of pollinators, beneficial insects, and pest cycles in ways that allow integrated pest management to replace chemical defaults.

What is important to notice is that the same AI technology being used for surveillance capitalism, predictive policing, and algorithmic engagement optimization is being used here for something structurally different. The technology has not changed. The conditions surrounding its deployment have. Farmers organizing their operations around soil health, biodiversity, and long-term ecosystem regeneration deploy AI in service of those goals. Industrial monocultures deploy AI in service of yield maximization at any ecological cost. The technology does not choose. The institutional context, ownership structure, and economic incentives surrounding the technology choose.

This is the book’s argument applied: intelligence is amplifying what lies beneath it. Regenerative agriculture using AI is not utopian. The systems are imperfect, the practices are partial, and adoption remains far below what climate stability requires. But the existence of the practice demonstrates that the same intelligence can be oriented in fundamentally different directions depending on what surrounds it.

Similar patterns exist across other domains. AI-driven climate modeling now operates at resolutions that allow regional governments to plan adaptation strategies that were impossible a decade ago. Disease surveillance systems built during the COVID pandemic have remained operational and are being adapted for early warning on antimicrobial resistance, vector-borne disease, and ecological tipping points. Scientific collaboration tools using language models are accelerating cross-disciplinary research at rates that would have been impossible under the older publication-based knowledge architecture. The biodiversity monitoring and collective deliberation work introduced in Chapter Six is part of this same pattern—intelligence being deployed in service of what extractive systems cannot perceive.

None of this is sufficient. Almost all of it operates against headwinds—extractive incentive structures, capital concentration in the labs producing the most capable models, regulatory frameworks that have not caught up to the technology, and the deeper civilizational fragmentation Part I diagnosed. But it is happening, and it is real, and it is the floor on which more ambitious work can build.