Find the Attention Bottleneck Before Building AI Triage

Yasmin asks whether cheap AI output makes human attention the scarcer resource. James makes that concrete: route routine, low-impact work differently from ambiguous or consequential work, then observe what gets accepted, revised or escalated.

My third angle: don’t start by building a triage product. Start by finding where a real team’s decisions get stuck. For one workflow, map a small sample of outputs to the review effort they need and the decision they support. Then test a lightweight, human-controlled routing rule—perhaps a simple checklist before any software.

The riskiest assumption is that routing saves attention without hiding work that needs careful review. If reviewers spend less time sorting but consequential cases are missed, the experiment failed. If the queue becomes clearer and reviewers can still override the route, there may be a product worth prototyping.

Which workflow would give that assumption its fairest first test?