Skip to main content
Large People ModelHuman Operating Architecture
AboutSearch the complete site3-minute readiness check
Back to Research & Evidence
Modeled estimatePublic record

The Supervision Ceiling

Why human oversight becomes the binding constraint as governed agents scale faster than supervisory capacity.

Grade A mechanism, modeled coefficientsJuly 13, 2026By Chad Stewart10 min read

Research record

Read the claim with its evidence boundary.

Evidence basis
Modeled estimate
Method
Framework synthesis with modeled assumptions described in the record.
Citations / endnotes
Formal citations or endnotes are not yet attached to this public record.
Version / date
Published July 13, 2026; a separate version identifier is not yet supplied.
Limitations
The evidence label and record sections define the current boundary. No customer outcome is inferred.
What would change the conclusion
Contradictory external research, benchmark data, or verified customer outcomes would require review.

Claim

Agent capability is not the only scaling constraint. Human oversight capacity is finite, and it can degrade before the dashboard shows obvious failure.

The strongest external support comes from human supervisory control, fan-out, cyber operations, vigilance, and automation-bias literature. The enterprise-agent coefficients remain unmeasured.

What the evidence supports

The structure of the claim is strong: one human cannot supervise unlimited autonomous work without loss of attention, situation awareness, or intervention quality.

  • Fan-out literature shows supervisory span can be estimated.
  • Automation bias is well established and named by the EU AI Act.
  • Effective human oversight by natural persons becomes a regulatory duty for new and materially changed high-risk systems from 2 August 2026.

What is still a hypothesis

The specific LPM coefficients for complexity, risk, reversibility, and base capacity are reasoned assumptions. They are not measured findings yet.

  • The low-override rubber-stamping band needs calibration.
  • The 1:5 comparison ratio comes from weak market sources and should be replaced with client-specific ratios.
  • The benchmark plan exists to validate or retire the coefficients.

Counterargument

Better interfaces, better alerts, and better workflow design may raise supervisory capacity. LPM should expect the ceiling to move as systems improve. The claim is not that capacity is fixed forever. The claim is that it is bounded and must be measured.

Keep the Evidence Clear

Move from research question to operating signal.

Use the related metrics, use cases, and knowledge objects to inspect the operating-model problem without overstating what has been proven.

Continue with evidence

01Research LibraryReturn to the complete collection of evidence records.02Metrics LibraryInspect the signals connected to this research question.03Resource HubContinue with practical guidance and working tools.