August 7, 2026
Human oversight that scales
The final 10 percent reveals where judgment, risk, and organizational trust still belong.
The group explored how leaders can increase capacity while preserving judgment, relationships, and organizational knowledge. The 90/10 problem showed why fast drafts can still require demanding expert work. Oversight should reflect consequence, reversibility, workflow maturity, and the organization’s adoption posture.
Questions the room worked
When does the final 10 percent require an expert?
How does oversight change as a workflow matures?
What helps teams trust AI-enabled work?
Methods you can use
Risk-based review
Match review effort to consequence and reversibility.
- 1.Low consequence and reversible: sample the output.
- 2.Moderate consequence: require a named reviewer.
- 3.High consequence or hard to reverse: require evidence and explicit approval.
The maturity progression
Move from assisted drafting to structured workflows, then monitored automation. Advance only when quality controls survive repeated use.
Trust through visibility
Show where AI contributes, where judgment remains decisive, and how errors are detected.
Four adoption postures
Plan for speed-first accelerators, repeatability seekers, quality or compliance resisters, and mixed organizations. Training, controls, and proof should change with the posture.
Progressive adoption
Begin with read-only reporting and audits. Move to templated work, external content, and connected workflows only after the earlier controls hold.
What carried forward
- Human review should increase with consequence and irreversibility.
- The final 10 percent may contain most of the judgment.
- AI creates capacity when the full workflow changes.
- Adoption varies by culture, role, and perceived job risk.
- Over-automating account notes can weaken human context.
- AI-created files may preserve metadata about their production trail.
- Clear limits and visible controls build trust.
What to test next
- Map one workflow by consequence, reversibility, and reviewer.
- Define the evidence required before an AI-assisted decision is approved.
- Track the work made possible by the capacity AI creates.
Take the work with you