More AI documents can mean slower decisions.

Video caption

More AI documents can mean slower decisions. The source exception is absent from the recommendation. Do this because document volume is a misleading proxy. #EricFieldNotes

Full written post / accessibility read

A company can multiply proposal, summary and ticket volume while the team spends longer discovering what was actually approved. Output throughput is not decision throughput. The missing metric is how often a commitment has to be reopened because its source or condition was lost.

Suppose a fictional leadership packet turns tentative customer interest into committed demand. A later agent drafts hiring and roadmap tasks from that claim. The team moves fast until sales opens the original note and reverses the assumption.

For each commitment, record whether the owner opened the primary source, which caveats survived, the action taken and any later reversal cost. A weekly sample reveals whether faster document flow is helping or hiding attention debt.

Let AI accelerate drafting, then invest human review at the conversion from proposal to commitment. Track time to an accepted decision and the cost of unwinding it. The goal is not fewer pages; it is fewer unsupported commitments.

#EricFieldNotes

Four-beat scene transcript

1. More AI documents can mean slower decisions.

A company can multiply proposal, summary and ticket volume while the team spends longer discovering what was actually approved. Output throughput is not decision throughput. The missing metric is how often a commitment has to be reopened because its source or condition was lost.

Visual: Draft count hides reversals and reconstruction work.

2. A polished packet can launch the wrong work.

Suppose a fictional leadership packet turns tentative customer interest into committed demand. A later agent drafts hiring and roadmap tasks from that claim. The team moves fast until sales opens the original note and reverses the assumption.

Visual: The source exception is absent from the recommendation.

3. Measure source readback and reversal.

For each commitment, record whether the owner opened the primary source, which caveats survived, the action taken and any later reversal cost. A weekly sample reveals whether faster document flow is helping or hiding attention debt.

Visual: Sample consequential decisions, not all generated prose.

4. Optimize decisions that survive.

Let AI accelerate drafting, then invest human review at the conversion from proposal to commitment. Track time to an accepted decision and the cost of unwinding it. The goal is not fewer pages; it is fewer unsupported commitments.

Visual: Do this because document volume is a misleading proxy.

Research and claim limits

• NIST AI RMF 1.0 (S171)
• Lee et al., The Impact of Generative AI on Critical Thinking, CHI 2025 (S172)

The examples identified as illustrative or simulated are design probes, not reported incidents. Vendor specifications do not establish workload performance.