Proof-Adjusted Autonomy: The 90% Agent Is a 61.6% Agent

Short summary

The author introduces Proof-Adjusted Autonomy (PAA), a metric showing that agents billed as 90% autonomous are actually 61.6% when you factor in evidence completeness, independent validation, and on-time delivery. PAA = P(A) × P(C|A) × P(R|A,C) × P(T|A,C,R), multiplying four conditional gates. The gap between raw and proof-adjusted autonomy becomes review queues, silent risk, and rework. Long-horizon agents need proof at step boundaries because reliability multiplies, not averages.

• •Proof-Adjusted Autonomy reveals 90% autonomous agents are really 61.6% production-ready
• •PAA multiplies four conditional gates: execution, evidence, validation, and timing
• •Reliability compounds — 99% per-step over 50 steps yields only 60.5% clean completion

Generated with AI, which can make mistakes.