The Snap
Why Intelligent Systems Know When to Stop Searching
The Mystery
For years, AI appeared to behave like a search engine.
More search.
More comparison.
More evaluation.
More reasoning.
If uncertainty remained, the obvious response was to keep searching.
But modern intelligent systems increasingly behave differently.
At some point, they stop exploring.
They stop comparing.
They stop recomputing.
They act.
The question is why.
The Cost of Continuing
Every additional inference has a cost.
Retrieve another document.
Compare another supplier.
Evaluate another possibility.
Verify another source.
Each step may reduce uncertainty.
But each step also consumes computation.
Early in the reasoning process, this trade-off makes sense.
Confidence rises quickly.
The cost is justified.
Eventually something changes.
Each additional computation produces less additional certainty.
The system reaches diminishing returns.
The Snap
There is a moment when further search no longer materially improves the expected quality of the decision.
But it still increases computational cost.
At that point, the economics change.
Continuing becomes more expensive than acting.
This is the snap.
Not a sudden leap in intelligence.
A transition in optimisation.
The system no longer asks:
“Could I search again?”
It asks:
“Do I need to?”
The Threshold
Perfect certainty almost never exists.
Waiting for complete certainty would require endless computation.
Intelligent systems optimise for something different.
Sufficient certainty.
Once uncertainty falls below the threshold required for reliable action, the search ends.
Not because every possibility has been explored.
Because additional exploration is no longer economically worthwhile.
The decision becomes computationally complete.
Why Defaults Matter
This explains the true role of defaults.
Defaults are often described as preferences.
Or habits.
Or shortcuts.
They are none of these.
Defaults are accumulated certainty.
They allow an intelligent system to reach the action threshold without recomputing every possibility from scratch.
Every successful reuse strengthens confidence.
Higher confidence reduces uncertainty.
Lower uncertainty means the threshold is reached sooner.
Defaults are not simply answers.
They are permission to stop searching.
Agency Begins Here
Every autonomous system faces the same decision.
Continue thinking.
Or act.
Without a stopping principle, intelligence never becomes agency.
It simply searches forever.
The snap is the moment reasoning becomes action.
The point where accumulated certainty outweighs the value of additional computation.
Every intelligent agent must eventually make this transition.
Beyond Resolution
The Resolution Principle explains why intelligent systems minimise uncertainty.
The Snap explains when they stop.
Together they describe a complete cycle.
Intent.
Uncertainty.
Resolution.
The Snap.
Action.
Memory.
Better priors.
Lower uncertainty.
Repeat.
As AI systems become more capable, this transition will become increasingly important.
The future may not belong to the systems that think the longest.
It may belong to the systems that know exactly when enough uncertainty has been resolved to act.