What to automate in hiring, and what to keep with your team.

Dana describes the role in a few sentences: the product, the customers the hire will own, and what the first year has to produce. Yardstick drafts the outcomes for the role and designs the interview guide around them: the questions, the follow-ups, and a scorecard with a written anchor for every score. Dana reads it, changes two outcomes that were aimed at a bigger company, and cuts one question.

Why: This is the step that matters most, because the job description, the application scores and the interview all come from it. AI drafts it so it gets done well and fast, and you read and edit it until it describes the job you actually need filled.

Dana’s own agent drafts the job description from the same outcomes, so the post and the interview ask for the same things. Dana edits the tone and publishes it to the careers page.

Why: It is a draft, and nobody outside the company sees it until Dana publishes it. Read it closely anyway, because it is the first thing a candidate sees and it should ask for the same outcomes the interview checks.

Application review scores every application against the criteria Dana set for the role. Dana works down the queue with the scores beside each name and picks the eight people she wants to talk to.

Why: The reading is volume work and suits AI. Choosing who gets an interview hour is a choice about a person, so it splits: the scores read, Dana chooses.

The built-in agent screen new applicants nightly reads each day’s applications against the role and prepares a rejection draft for the ones that do not match, held for review. Dana confirms the ones she agrees with, and those drafts move into her normal review before they go out.

Why: The email reaches a candidate, and a wrong one costs a good applicant. The agent writes it; a person releases it.

Interview times come from the Cal.com or Calendly account Dana already uses, and her own agent drafts the invitation for her to send.

Why: The invitation goes to a candidate, so a person sends it. The back and forth over times is what the calendar tool is for.

The built-in agent prep interview kits for next week’s sessions emails each assigned interviewer their kit: the candidate and role context, the questions, the scorecard anchors, and the logistics.

Why: It goes to Dana's own team, it only reads the schedule and candidate records, and a kit with a gap in it gets noticed by the interviewer who opens it.

Dana and her head of support interview from the guide. The follow-ups are written into the guide, and which one to ask depends on what the candidate just said. That is also what interviewing when candidates are using AI comes down to.

Why: The interview is a judgment about a person, made in real time, and it is where the evidence comes from.

Each interviewer scores the candidate against the written anchors before seeing anyone else’s score, so two people who heard the same answer can land on the same number. Why interviewers disagree covers what happens when they do not.

Why: The score is the interviewer's record of what they heard. It is only evidence if the person who heard it wrote it.

The built-in agent chase missing scorecards sends reminders to the interviewers who have not filed feedback, and the last reminder escalates to Dana as the hiring manager.

Why: It is the most repetitive job in hiring, it goes to the team, and a reminder sent one day early costs nothing.

Once the assigned scorecards for a candidate are in, the built-in agent prepare a panel decision brief writes one brief from the completed scorecards for the panel to review.

Why: It is a summary of evidence people already wrote, read by the same people. The debrief starts from the scorecards instead of from memory.

Dana and her head of support read the brief, talk through the one criterion where they scored differently, and choose. Dana calls the candidate with the offer herself; the job offer guide covers that call.

Why: This is the step every other step exists to inform.