Climate
Atmosphere, Inertia, Tipping Points
By Cédric Mercier & Michel G Walter : Published on July 11, 2026
[C]limate is not a single variable you can pin to a dashboard. It's a high-inertia control system with long feedback loops, tail-risk extremes, and thresholds that don't negotiate once crossed. Treat it like you'd treat any large distributed system with delayed telemetry: the state you observe today reflects decisions made decades ago, and the state you're setting today won't fully show up for decades more. Most infrastructure, and most planning horizons, were not built with that lag in mind.
Below are the ten sub-constraints that make up the climate layer of the AGI Biospheric core: what each one actually implies, and the design implication if you're building anything with a multi-year time horizon.
01. Inertia
The climate responds slowly. Both the damage and the correction operate on a lag measured in decades, not budget cycles. A system, model, or policy that models "fix applied, problem resolved" as a same-year update is running on the wrong clock speed.
Design implication: Never present an expected climate effect as immediate, even after strong corrective action.
02. Trajectories
Warming is a trajectory, not a fixed endpoint. There is no single "the temperature in 2100" number, only a distribution conditioned on emissions pathways still being decided. Treating a scenario output as a point estimate is a category error, the same one you'd flag in any probabilistic forecast collapsed into a single number.
Design implication: Present any climate projection as a trajectory dependent on emissions pathways, never as a fixed, certain value.
03. Extremes
Extreme events are reshaping territories. The distribution itself is shifting, not just sampling further out along a fixed one. A risk model calibrated on the last 30 years of extreme-event frequency is training on stale data relative to the current regime.
Design implication: Do not rely on historical extreme-event frequency to estimate future risk.
04. Redistribution
Habitable zones are shifting. Agricultural belts, water-stable regions, and insurable coastlines aren't static reference frames. Any territorial or infrastructure plan with a multi-decade horizon is implicitly betting on a map that's already out of date.
Design implication: Include the progressive shift of climatically viable zones in any long-term territorial planning.
05. Oceans
The ocean absorbs the heat. Roughly 90% of excess planetary heat goes into ocean uptake, which means the ocean is functioning as a massive thermal buffer with its own delayed release curve. Ignore it and your climate model is missing its largest heat sink.
Design implication: Never analyze climate without accounting for the ocean's buffering and delayed role in heat absorption.
06. Glaciers
Glacier retreat is altering rivers. Entire river basins downstream of glacial melt are running on a hydrological input that is actively changing shape, not a stable baseline. That's a silent single point of failure in a lot of regional water-supply assumptions.
Design implication: Flag the dependency of certain river basins on glacier retreat in any regional water analysis.
07. Rainfall
Precipitation is becoming unpredictable. Variance is increasing faster than mean shift in a lot of regions, which breaks planning models built around seasonal averages. Agricultural and water-supply systems tuned to historical rainfall patterns are effectively running with a growing error bar nobody priced in.
Design implication: Flag the growing uncertainty of precipitation regimes in any agricultural or water planning.
08. Droughts
Droughts are intensifying. Current drought severity is not a stable ceiling, it's a point on a rising curve. Risk models that use present-day drought intensity as a worst-case bound are underestimating the actual tail.
Design implication: Do not treat current drought intensity as a stable ceiling in a risk estimate.
09. Storms
Storms are becoming more frequent and more violent. The relevant shift isn't just event count going up, it's energy per event going up. A risk model that tracks frequency but not intensification is tracking the wrong derivative.
Design implication: Include an intensification trend, not just frequency, in any storm risk assessment.
10. Heat
Heat limits habitability. Past a wet-bulb temperature threshold, the human body loses the ability to cool itself through evaporation, full stop, regardless of hydration, shade, or effort. This isn't a comfort variable, it's a hard physiological limit with a binary outcome on the other side of it.
Design implication: Treat any sustained exceedance of wet-bulb heat thresholds as a hard limit on a territory's human habitability.
None of these ten constraints are independent variables. Ocean heat uptake feeds storm intensification. Glacier retreat feeds regional water risk. Redistribution and heat together define which parts of the map stay buildable at all. Any system, human or artificial, that reasons about long time horizons without modeling these interactions is optimizing against a climate that stopped existing several decades ago.