---
name: capability-saturation
description: Estimate when adding more AI agents stops improving task performance because coordination cost and duplicated work dominate. Use for multi-agent team sizing, orchestration planning, or reviewing a proposed agent count.
---

# Capability Saturation

Estimate a practical agent-team size from observed solo reliability, task overlap, and coordination overhead.

## Workflow

1. Gather the single-agent success rate from representative tasks.
2. Estimate overlap: the share of work likely to be repeated across agents.
3. Estimate coordination overhead per additional agent, including handoffs and reconciliation.
4. Compare team sizes from one through twelve using:
   - coverage gain from additional partially independent attempts;
   - a coordination penalty that compounds with team size.
5. Identify the highest-scoring team size and report the nearby alternatives.
6. Treat the result as a planning estimate, then validate it with a small benchmark.

## Output

Return the suggested agent count, estimated peak score, selected-team score, assumptions, and one recommendation for measurement.

Never present the estimate as a universal scaling law.

## Research basis

[Capable language models can outgrow the benefits of collaboration](https://doi.org/10.1038/s42256-026-01268-y) — Kim et al., *Nature Machine Intelligence* (2026). This peer-reviewed study describes the capability-saturation effect in multi-agent systems.