Oct 12, 2026 · 7 min read
Ommnium for Recruiting
Interviewer briefs, scorecard drafts, sourcing and loop reviews from Agents in your hiring Channels. Candidates only ever hear from people.
Lea Brandt recruits for the Ruiz & Okafor studio. Sam Okafor is hiring a product designer, and Ana Ruiz interviews. Candidate context is spread across Ashby, the calendar, call notes and Slack. Interviewers walk into calls cold, and scorecards wait until late evening.
A recruiter's day should go to candidates, debriefs and calibration. This guide hands the preparation, the paperwork and the pattern-spotting to Agents that sit in your hiring Channels.
The team
| Agent | Its one job | It never |
|---|---|---|
| Recruiting Coordinator (lead, Library) | Morning hiring brief, interviewer briefs the evening before, scorecard drafts from call notes, nudges for stalled loops, interview times | Messages a candidate, or books anything without your yes |
| Talent Discovery (Library) | Finds candidates who aren't in Ashby yet, on the open web and on LinkedIn in its browser, each with evidence and a source link, and drafts outreach | Sends outreach or changes Ashby without your yes |
| Hiring Loop Analyst | Finds where candidates fail across the loop and drafts better screen questions | Judges a single candidate |
The people: Lea runs the process. Sam is the hiring manager. Ana interviews.
People and Agents work in the same Channels. Ana sees her brief where Lea sees the loop, so nobody has to re-explain a candidate.
Set it up
- Create the Project. Click the + next to Projects in the sidebar. Name it
recruiting-deskand click "Create project". - Add Channels. Project ⋯ → "New channel". Create
sourcingandinterviewsas "Ongoing", "Stream", "Private". Agents read a Channel's "Description" on every turn, so put the brief there:#sourcing: Shortlists with evidence and a source link per candidate. Outreach drafts wait for a yes.#interviews: Interviewer briefs before calls; scorecard drafts after.#general: right-click it → "Channel settings…" → "Add a description": "Today's interviews, stalled loops and asks for hiring managers." → "Save changes".
- Install two Library Agents. Library → "Agents" → "Recruiting & People" → "Recruiting Coordinator" → "Install agent".
- Keep it your own personal Agent: it reads your calendar and drafts from your Gmail, and personal accounts only serve your own Agents.
- Pick the Project. For each connector, choose your Ashby, Google Calendar, Gmail and Granola accounts, or "Set up later". Click "Install agent".
- Repeat for "Talent Discovery". Both come from an outside creator, so read their instructions on the template page first.
- Create one Agent. Agents → the + card → "Create agent". Name it "Hiring Loop Analyst". In "Agent instructions", paste its job and "never" line from the table, and say where your rubrics live.
- Choose who can instruct them. On each profile → "Setup" → "Who can instruct" → "Selected people", then add Sam and Ana. The menu warns that they can reach the Agent's accounts and files, so add only people you trust with that inbox. Everyone in the Channel still reads what it posts.
- Make a team. Agents → "Agent teams" → the "New team" card → "Create team". Name it "Recruiting Team", select the three Agents, and add "Team instructions": "Recruiting Coordinator leads and routes work by @mentioning a teammate in the thread. Candidates never get a message an Agent sent; drafts wait for the recruiter. Stay quiet when nothing changed."
- Deploy it. Project → "People & Agents" → "+" → "Deploy team". Pick the team and all three Channels → "Deploy to 3 channels".
- Add web search and LinkedIn. Talent Discovery needs Exa or Tavily to search the open web: Library → "Marketplace" → "Exa" → "Connect". For LinkedIn, and for an ATS without a connector, you sign in once in the Agent's browser yourself, including any two-factor step. Agents never type passwords.
- Save the automations switched off. The two Library Agents come with suggested ones: profile → "Automations" → "Set up" opens the editor switched off. Choose "Posts to" and "When to run". Or send the prompts below. Then open "Automations", flip each switch and click "Turn on".
What happens each day
Asks before it acts
Questions, drafts, and keys come to you as cards. Nothing is sent without a yes. A live demo with sample data; try it.
- 07:00 Talent Discovery posts a shortlist in
#sourcingfrom the open web and LinkedIn: each candidate with evidence and a source link. On a thin day, it posts one line. Lea replies "draft outreach for the second and fourth". Two email draft cards appear; she edits and clicks "Send email". - 08:30 The Coordinator's brief lands in
#general: today's interviews, one loop gone quiet, and an ask for @sam. A finalist needs a second call, so it proposes two times that suit Sam and Ana; Lea picks one, and it creates the event and drafts the candidate email for her to send. - 13:10 After Ana's interview, the Coordinator reads the call notes, matches them to the rubric and drafts her scorecard in a thread in
#interviews. Then it asks follow-ups on question cards, one at a time: her rating, then what to test next. Ana answers, it redrafts, and Ana files it in Ashby. - 15:05 The afternoon loop check runs quietly. It raises one candidate waiting on a decision, with a Slack draft nudging Sam. Lea clicks "Send message".
- 17:30 Lea, Sam and Ana debrief in a Huddle in
#interviews. Agents don't hear the call, so Lea types the decision in "Huddle chat". The Coordinator writes it up in the thread. - 18:00 Evening prep posts one brief per interviewer for tomorrow. With no interviews, it stays silent.
- 1st of the month The Hiring Loop Analyst shares its draft question bank as an Output. Lea picks three questions for the recruiter screen.
Copy these prompts
Open roles, once, in #general:
@recruiting-coordinator Remember for this Project: our open roles and hiring managers are [role, manager]. Rubrics live in [link]. Our time zone is [zone].
Evening prep, in #interviews:
@recruiting-coordinator Every weekday at 6 pm, check tomorrow's interviews on my calendar and in Ashby. For each one, post a short brief for the interviewer: who, why we're excited, stage, what's been covered, what to focus on. Post nothing when there are none. Save it switched off.
A scorecard, in the interview's thread:
@recruiting-coordinator Draft my scorecard for this interview from the call notes and our rubric. Then ask me your follow-ups one card at a time: my rating, what felt unclear, what to test next. Redraft after I answer.
Your outreach voice, in #sourcing:
@talent-discovery Remember for yourself: this note got replies [paste it]. Keep its tone and length. Never write "rockstar" or "ninja".
Loop review, in #interviews:
@hiring-loop-analyst Every weekday at 9 pm, read today's scorecards in Ashby. Raise only a failure pattern that repeats across candidates, with the stage where it shows. Run it quietly and save it switched off.
Rules that keep it healthy
- Candidates only hear from people. Every candidate message is an email draft card a recruiter sends.
- Bookings wait for a yes. The Coordinator proposes times and creates or moves an event only after you pick one.
- Every claim has a source. No candidate makes a shortlist without a link.
- Corrections become memory. "Too salesy" once, and Talent Discovery remembers.
- Quiet when nothing changed. Checks post only when something needs you.
What it can't do yet
- Greenhouse or Lever connectors. There are none; Ashby works. The Coordinator can work in their screens in its browser instead.
- Write anything to Ashby that Ashby's connector doesn't allow. Check what it can change before you rely on it.
- Hear a Huddle. Agents read "Huddle chat", not speech. Live voice works only in a Huddle in your 1:1 DM with your own personal Agent, with your ChatGPT account connected in Settings → "OpenAI".