Budget human attention like review capacity.

Video caption

Budget human attention like review capacity.

AI summaries can conceal unread source material.

Define a trigger and a short source readback.

My rule: Do this because throughput is an end-to-end property.

#EricFieldNotes

Full written post / accessibility read

A team may generate plans and recommendations faster than leaders can examine their sources. The answer is not asking everyone to read everything. It is allocating explicit review time to decisions with material customer, financial or architectural consequences.

A manager approves a queue of polished plans but has not seen the condition that limits one customer rollout. More document flow creates the appearance of throughput while a later reversal consumes several teams. This is a plausible mechanism, not a measured industry rate.

For rights, money, safety, external commitments or core architecture, route the decision to a named owner with the authoritative passage and uncertainty. Ask them to restate the condition before approving the next action. Audit the readback on a small sample.

Keep drafting fast, but count decisions that survive implementation and customer feedback, plus reversals and reconstruction hours. That tells a leader whether AI improved decision capacity or merely increased the pile waiting to be understood.

#EricFieldNotes

Four-beat scene transcript

1. Budget human attention like review capacity.

A team may generate plans and recommendations faster than leaders can examine their sources. The answer is not asking everyone to read everything. It is allocating explicit review time to decisions with material customer, financial or architectural consequences.

Visual: AI accelerates document supply; decisions remain scarce.

2. The queue becomes invisible.

A manager approves a queue of polished plans but has not seen the condition that limits one customer rollout. More document flow creates the appearance of throughput while a later reversal consumes several teams. This is a plausible mechanism, not a measured industry rate.

Visual: AI summaries can conceal unread source material.

3. Create a consequence-based review lane.

For rights, money, safety, external commitments or core architecture, route the decision to a named owner with the authoritative passage and uncertainty. Ask them to restate the condition before approving the next action. Audit the readback on a small sample.

Visual: Define a trigger and a short source readback.

4. Measure accepted decisions per review hour.

Keep drafting fast, but count decisions that survive implementation and customer feedback, plus reversals and reconstruction hours. That tells a leader whether AI improved decision capacity or merely increased the pile waiting to be understood.

Visual: Do this because throughput is an end-to-end property.

Research and claim limits

• NIST AI RMF 1.0 (S171)
• Lee et al., The Impact of Generative AI on Critical Thinking, CHI 2025 (S172)
• Ocasio, Towards an Attention-Based View of the Firm (S81)

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