The Sweet Spot

Hyperbolic Reduction

When you're trying to evaluate a decision or approach, take it to the extreme — multiply it to 1000, then reduce it to 1. Analyse the hypothetical effect at both ends. This forces you to see the strengths and weaknesses of the approach at scale and at its simplest.

For example: what would happen if I had 1000 AI agents running multiple tasks versus only 1 agent running multiple tasks? With 1000 agents, you get massive parallelism but also massive coordination overhead — agents might duplicate work, conflict with each other, or produce inconsistent outputs. With 1 agent, you get simplicity and consistency, but it becomes a bottleneck that can't scale. The optimal answer is almost never at either extreme — the hyperbolic reduction helps you find the tensions that reveal the sweet spot.

Finding the sweet spot

Ask what breaks first at each extreme. At 1000: what's the first thing that fails — coordination, cost, quality control? At 1: what's the first thing you miss — speed, coverage, resilience? The sweet spot is where you maximise the benefit you'd miss at 1, without hitting the failure you'd see at 1000. You can narrow it down by halving from the top (1000 → 500 → 250) until the coordination problems disappear, and doubling from the bottom (1 → 2 → 4) until the bottleneck eases. Where those two meet is roughly your sweet spot — like a binary search for the right scale.

Testing for the sweet spot

Start small, measure, then scale incrementally. First, define your metric — what does "better" look like? Speed? Accuracy? Cost? Pick 1-2 that matter. Run the extremes first: test with 1 and with a high number (say 10-20) to get your floor and ceiling. Then binary search the middle — if 10 agents coordinate well but 1 is too slow, try 5, then adjust. Watch for diminishing returns: plot your metric against the number and look for the inflection point where adding more stops helping. Finally, stress test the winner — push edge cases, simulate failures, and make sure it's robust, not just optimal in ideal conditions.