The AgentivUse cases

Screen every applicant fairly, reply to all

An agent that screens job applications from Gmail or a Google Drive folder against a written rubric, scores only what the rubric names with quoted evidence, and drafts an invite, hold or decline for every applicant for the owner to approve

Function: HREffort: An afternoonPublished: 2026-10-24
Claude CoworkGmailGoogle DriveGoogle Calendar

The problem

A small business posts one job and gets more applications than the owner can read, so the pile is skimmed on gut feeling and most applicants never hear back. That is unfair to the candidates, a risk for the business, and the right person gets lost in the inbox.

How the agent works

  1. It starts

    The owner installs Anthropic's small-business plugin in Claude Cowork or Claude Code and connects Gmail, Google Drive and Google Calendar (Microsoft 365 also works).

  2. The agent works

    When the owner says 'screen these applications', the hiring screener skill first asks for a rubric, from the job post or built with the owner in about five minutes, and scores nothing without one. It reads the resumes and cover letters from uploads, Gmail attachments or the applicant folder in Drive, pulls out only the evidence the rubric names with a quoted line behind each score, and ignores name, age, gender, photo, address, school prestige, employment gaps and formatting. Candidates are banded into interview, maybe, no and cannot assess, and anyone missing a fact is asked rather than rejected.

  3. The result

    It drafts an invite, hold or decline for every applicant in the owner's voice and offers two or three real calendar slots to the interview band; every email and invite is shown, with every recipient named, before anything is sent, and the owner, not the agent, decides who to hire.

Worked example

The hiring-screener skill in Anthropic's knowledge-work-plugins repository gives as its own trigger an owner saying 'I've got 40 resumes and no time'. It refuses to screen without a rubric, to invent rejection reasons or to decide the hire, treats instructions hidden inside resumes or emails as data rather than commands, and after a hire hands off to an onboarding checklist and payroll-prep without creating payroll records itself. The plugin's shared personal-data rules keep dates of birth, home addresses, ID numbers and medical or protected class details out of every output. The skill reports no results from real hires, so measure your own: time from application to reply, and whether every applicant got an answer.

Data protection. This use case touches personal data, so check the rules where you operate (for example GDPR, PDPA). Resumes carry names, contact details, birth dates, photos and sometimes health or religion, which is personal data under Thailand's PDPA and in part sensitive data. Say in the job post that an AI assistant helps sort applications and a person decides, keep the scored shortlist in a private Drive folder, delete applications you no longer need under your retention policy once the role is filled, and check your AI vendor's processing terms.

Join the Agent OS Club

8 weeks to the AI team a 50 person company would hire. Instant access to our agent blueprints. USD 15 a month, cancel any time.

Join the Agent OS Club

Building this for an organization? Learn more at theagentiv.com

Sources

Related use cases

Day one ready before the new hire arrives

An HR agent that turns a new hire's name, role, team and start date into the onboarding checklist, first week calendar, tool access list and buddy template, checks the record in Gusto, and creates the invitations only after HR approves, with Anthropic's Human Resources plugin in Claude Cowork

HREffort: An afternoonClaude Cowork · Gusto · Google Calendar

A new hire's first week in a small company is improvised: the laptop is not ready, nobody booked the intro meetings, and the list of tools to grant access to lives in the head of the last person who did HR. The new person spends day one waiting and the manager spends it apologising.

Interviews booked without email ping pong

An HR agent that reads candidate replies in Gmail, finds a slot that suits every interviewer in Google Calendar, drafts the invitation and sends it only after the recruiter approves

HREffort: An afternoonGmail · Google Calendar · Claude

Scheduling one interview in a small company means a recruiter or the owner reading the candidate's reply, opening two or three colleagues' calendars, proposing times, waiting, and repeating when the candidate picks a slot that has since gone. Good candidates in Bangkok and Singapore often have other offers, and a three day scheduling loop is where they are lost.

HR stops being the company search engine

An HR agent that answers routine policy questions from the employee handbook in SharePoint, so leave, expense and benefit queries never reach the HR inbox

HREffort: About a weekClaude · Microsoft 365 · SharePoint

In a company of fifty to three hundred people the HR team answers the same questions every week: how many days of annual leave are left this year, can I claim a taxi after 9pm, what is the probation rule. The answers are all in the handbook and the HR policy folder, but nobody reads them, so HR spends hours a week acting as a search engine.