scarcity migrating upward through the stack my thesis for the next few years is that the most interesting consequence of abundant intelligence might be where scarcity moves next technology rarely eliminates scarcity - it moves it abundant compute makes intelligence cheaper abundant intelligence makes good judgement scarcer abundant judgement makes trust scarcer abundant autonomous systems make authority scarcer compute → intelligence → judgement → trust → authority each solved bottleneck creates the next one above it compute produces intelligence intelligence produces judgements judgement earns trust trust earns authority authority determines what gets to affect reality this means the end state of cheap intelligence could look less like an intelligence problem and more like a giant control problem we get more models, agents, outputs and possible actions than humans can meaningfully supervise - so the bottleneck moves toward deciding what deserves resources, context, trust and ultimately permission to act but autonomy pushes this another level deeper evaluation can no longer be something done before deployment. once agents continuously operate in changing environments, judgement has to enter the execution loop itself - deciding whether an action was good, what should be trusted, what should be reinforced and what should happen next evaluation stops being QA and becomes part of the control plane and eventually even evaluation has its own problem - who evaluates the evaluators? when thousands of models, agents and humans can produce competing assessments, judgement becomes abundant in quantity while remaining scarce in quality the problem shifts from generating judgement to discovering which sources are actually calibrated - not who sounds convincing, but who repeatedly assigns the right confidence to reality and updates correctly as it changes and the further you move through this stack, the less purely technological the problem becomes compute can be allocated by a scheduler trust, capital and permission to act require mechanisms for deciding which competing claims about reality deserve consequence authority without accountability creates its own bottleneck - once agents can affect reality, systems also need to determine who stands behind an action when its wrong incentives, reputation, collateral and mechanisms for discovering reliable judgement start looking less like peripheral features and more like infrastructure the deepest scarce resource may simply be permission to affect reality an autonomous economy cant let every intelligent system act - it needs mechanisms that continuously determine what gets trusted, what gets capital, what gets reinforced and ultimately what gets to act which brings me back to the point above - AI may commoditise intelligence without commoditising consequence instead it could make the systems deciding which intelligence gets to affect reality increasingly valuable so the biggest infrastructure opportunity around abundant intelligence may not be producing intelligence at all it may be building the systems that decide which intelligence deserves consequence and if scarcity migrate upward through the stack, value should migrate with it which is why im increasingly interested in the layers above intelligence rather than intelligence itself scarcity migrates → value follows → eventually the valuable infrastructure governs consequence rather than produces intelligence