O*NET work tasks · task-11280
Compose letters in reply to correspondence concerning such items as requests for merchandise, damage claims, credit information requests, delinquent accounts, incorrect billing, or unsatisfactory service.
An atomic unit of work as agent-native research objects: the O*NET task, its measured AI-usage and model-exposure figures (each grounded in its named source), and the collapse/survive judgment — separated by their non-collapsing four epistemic layers. An O*NET fact / a measured figure is {T,T,T,T}; the judgment is occurred=false AND adjudged=false. The same agent-native research-object format as the rest of the Atlas Commons. CC0.
"Compose letters in reply to correspondence concerning such items as requests for mercha…" is an O*NET task for Correspondence Clerks
SourceCitation
O*NET lists "Compose letters in reply to correspondence concerning such items as requests for merchandise, damage claims, credit information requests, delinquent accounts, incorrect billing, or unsatisfactory service." as a task for Correspondence Clerks (SOC 43-4021), importance 3.78.
How to read this claim
Superseded when O*NET retires or revises the task (reviewStatus -> superseded). Verbatim official fact.
21.7% of measured AI conversations for Correspondence Clerks map to this task (Anthropic)
SourceCitation
The Anthropic Economic Index measures 21.7% of AI conversations for Correspondence Clerks mapping to this task (dominant pattern: task iteration). A measured share of usage, NOT a percent of the job.
How to read this claim
A measured figure from a named release; superseded by the next AEI release. occurred=TRUE (it was measured), unlike the collapse judgment below.
LLM task exposure (Eloundou et al., gamma) for this task is 1
SourceCitation
The Eloundou/OpenAI GPTs-are-GPTs study estimates an LLM exposure (gamma) of 1 (E1) for this task — a model-rated codifiability estimate.
How to read this claim
adjudged=TRUE: peer-reviewed academic estimate. A model-rating, not a measure of jobs lost.
Whether this task collapses to AI or survives is an interpretation, not a fact
Interpretation
AgenticU's reading (exposure: E1): the codifiable share of this task overlaps with AI; what survives is the grounded-trust / human-becoming remainder. Advice over the data above, not a measured fact or an official verdict.
How to read this claim
occurred=FALSE (a reading, not an event) and adjudged=FALSE (AgenticU's view, not an authority's verdict) — the lowest-warrant claim. A flat truth-bit would render it identically to the O*NET fact; the Commons does not.