Hiring when agents write the first draft

A sales or support rep's day used to include composing replies from a blank page, over and over, for every incoming thread. When an inbox agent drafts a first response and queues it for review, that task is mostly gone. What replaces it is a faster, higher-frequency task: reading a drafted reply, deciding in seconds whether it is accurate, on-tone and complete, and then sending it as it is, editing it, or rejecting it and writing something different.

That is a real skill. It is not the one a job posting asking for "strong written communication" usually tests. Composing from scratch tests whether someone can build a message well. Judging a draft tests whether someone can spot what is wrong with a message quickly, under time pressure, across a high volume of them. That draws on a different kind of attention: pattern recognition and error-spotting, not generative writing.

Someone can be a strong writer and a slow, uncertain editor of other people's drafts. The reverse also happens: a fast, accurate editor who would take much longer to produce the same quality from a blank page. Neither pairing is unusual, because writing and judging writing draw on related but distinct abilities. An interview built entirely on a writing sample, "draft a cold email to this persona", measures the first skill and says little about the second. Yet the second is what the job requires hour to hour once an agent does the first-draft work.

The risk of testing the wrong skill is not abstract. A candidate who interviews well on a drafting exercise may turn out to rubber-stamp agent drafts without catching real errors. Approving quickly under volume is a different discipline from writing carefully once, unhurried, in an interview. The reverse risk exists too: a strong editor passed over because a drafting-focused interview never showed the skill that predicts success in the role as it is now structured.

The fix is to test the actual task. Give a candidate a set of already-drafted responses, some good and some subtly wrong: missing context, wrong tone, an inaccurate claim. Ask them to decide what to do with each one and explain why, under a time limit that resembles real working conditions. That measures judgment speed and accuracy directly. It does not infer them from how well someone writes with unlimited time and no draft to react to.

SalesCrew's Inbox agent ships in review mode by default. It drafts a reply and places it in a queue rather than sending it. A person using SalesCrew day to day is largely doing the judgment task described here: reading a drafted reply beside the evidence behind it, the thread and a confidence signal, and deciding whether to approve, edit or reject. Hiring and onboarding for a role that will spend real time in that queue should test for what the queue demands. Not the drafting exercise that used to be the job before an agent did the first pass.