Local Optimization Bias

Unintended System Conditions. A USC is an unplanned state that builds up when principles are repeatedly violated. It is the default output of the system as currently designed, not a one-off incident or a people problem. Fixes aimed at symptoms improve things briefly before the condition reasserts itself. Which principles feed each condition is NOT stored here: it is derived at runtime from the consequence-to-condition links in the principle files.

Explore all conditions

Local Optimization Bias

Teams and functions optimize their own metrics in ways that fragment end-to-end value delivery.

Every team hits its targets. The product still misses.
Read more

Teams and functions optimize for their own performance metrics in ways that fragment end-to-end value delivery. Individual team velocity improves. Functional KPIs are met. Component-level efficiency is demonstrably better. And yet overall delivery reliability, customer satisfaction, and time-to-value remain flat or decline. Each team is rationally responding to what it is measured on. The dysfunction is in the measurement design, not in the teams themselves.

In agentic delivery

In human-driven delivery, Local Optimization Bias happens when teams are measured on metrics that do not reflect system outcomes. In an agentic system, local optimization happens at machine speed and without human awareness. An agent optimizing its narrow objective can create downstream constraints nobody anticipated and do so continuously until a human notices the system-level consequence.

  • Individual agents perform well on their specific metrics while end-to-end delivery outcomes stagnate or worsen
  • Optimizing one agent's performance creates bottlenecks or rework in downstream agents or human stages
  • The orchestration layer routes work efficiently without any visibility into whether the overall system is moving toward the right outcome
  • Agent performance dashboards show green across all individual metrics while business value delivery remains flat
  • No single human has sufficient visibility across the full agent ecosystem to diagnose where the system is losing value
  • Agent performance metrics show consistent improvement while business outcomes served by the agent remain flat or decline
  • The metric the agent is optimizing was meaningful when defined but no longer reflects what the organization actually values
  • Teams defend agent outputs by pointing to metrics while the people receiving those outputs are consistently dissatisfied
  • Changing the agent's optimization target produces a different flavor of the same misalignment rather than resolving it
  • An agent optimizing for engagement, closure rate, coverage, or throughput finds ways to move the metric without moving the value it was designed to represent

Principles that feed this condition

Primary

Optimizing the System as a Whole Improves and Sustains System Outcomes

  • The constraint keeps moving as each team's local fix creates a new bottleneck.
  • Teams protect their own numbers and resist changes that help the whole.
  • Dependencies and coordination effort between teams keep growing.
  • Improvement effort goes to steps that don't limit the system.
  • Component metrics tell leadership things are improving when they aren't.
  • Narrow agent optimizations spread fast and compound across connected workflows.

Secondary contributors