Workplace

Recruiting

Not tested yet: Eric Siu's video shows the full setup; we read only the post text. Recheck the video for criteria and approval steps. HRBP is a listing from a recruiting software company; we haven't installed it. Recruiting Coordinator and Talent Discovery listings don't show clock times; you set them during setup. No recruiting listing we read states what a routine does on an empty day.

Grok Bot for recruiting: inbound screening on fixed criteria and a pipeline stall watch

How a founder runs talent Bots against the same hiring criteria every time, how a recruiting CTO has a Bot find where candidates stall, and what to keep human.

Work value 3/10

Our first recruiting playbook covers interview prep, scorecards and outbound sourcing. This one covers the other end of the funnel: candidates who apply to you, and the ones who get stuck between stages. It is for a recruiter, a hiring manager, or a founder doing their own hiring.

Not tested: The Daily Bot has no ATS or candidates, so we haven't run these setups. Everything below comes from the linked posts and docs.

What recruiting uses it for#

  • Screening inbound on the same rules every time. Eric Siu, founder of the ad agency Single Grain, posted on Sep 21: "We're using Juicebox and Grok Bot talent bots to make recruiting more repeatable." He lists what the system handles: "Inbound candidates," "Hiring criteria," "Candidate interest," "AI evaluation questions" and "Shortlisting." The point, he says, is "to apply the same criteria every time and spend human attention on the strongest candidates."
  • A concrete criterion. Siu gives one: "two promotions across two companies with 3-year average tenure." Writing criteria that specific is what lets a Bot check them.
  • Finding where candidates stall. Jony Musky, co-founder and CTO of a recruiting software company, describes a Bot called HRBP that "checks pipeline health, shows where candidates stall by stage, tells you when to push the search (sourcing, multipost, AI screens), and shortlists who is actually worth interviewing." He sells recruiting software, so read it as a vendor's description.
  • Running recruiting loops. Kevin Niparko, a PM at SpaceXAI, lists a "Recruiter: finding undiscovered talent and managing recruiting loops" on his own Bot team in Grok Bot for PMs.

Connectors to set up#

Connect your ATS, the Gmail inbox where applications arrive, a Google Sheet or Notion page with the written criteria, and Slack for the hiring channel. Add them from Connect apps in the sidebar (Cursor's docs). If your ATS has no plugin, the Bot can sign in on its own computer, and you take over for the password and two-factor steps (Take over for sensitive steps).

Siu runs the criteria through Juicebox, which he says "also coordinates the agents running each step." You don't need it to copy the pattern. A shared criteria doc works with any ATS.

Routines#

No source here publishes a clock time. Siu's broader advice for every Bot is "Keep a human in the loop until the bot earns more freedom," from his Sep 2 post. So:

  1. Run the example prompt by hand on yesterday's applicants for one role. Compare its Shortlist with yours.
  2. Tighten the criteria doc wherever you disagreed. Then ask the Bot to save the method as a skill (Save a skill).
  3. Add the weekly Pipeline stall check from the routine above, with a quiet rule. Click Test on it first (Routines).

Example prompt#

This is our prompt, built from Siu's five steps. It stops before anyone is moved or contacted:

Prompt
Read the hiring criteria doc for this role. Review every new inbound applicant in the ATS since yesterday. For each one, check the criteria one by one and quote the line in their resume or profile that meets or misses it. Sort them into Shortlist, Maybe and No with one line on why, and list anything you couldn't verify. Draft a short reply for the Shortlist only. Don't move anyone in the ATS, don't reject anyone, and don't contact any candidate.

What to expect#

These are self-reported, and none comes with a time figure or a hiring result. Siu's claim is consistency: the same criteria applied to every candidate. Musky's is visibility: which stage is the bottleneck this week.

Gotchas#

  • Candidate privacy. Cursor's Talent scout starter says to "Keep external outreach behind approval, and have the Bot respect candidate privacy and each source's terms."
  • Criteria can encode bias. A rule like Siu's tenure signal screens people out at scale. Have a person review the No pile, and check your criteria against your local hiring law before you automate them.
  • Rejections stay human. The example prompt drafts replies for the Shortlist only. Siu says most of his Bots "cannot automatically send, publish, or change live pages."
  • One job per Bot. Siu: "We made the mistake of trying to build one super-agent that could do everything. It became harder to manage, harder to evaluate, and less reliable." Keep screening and the stall check in a recruiting Bot that does nothing else.
  • Busy inboxes cost usage. Each run spends usage, and a frequent schedule "can use a week of usage in a day" (Routines). Daily screening is usually enough.

Sources#

Grok Bot for recruiting: interview prep, scorecards, and sourcing batches

How SpaceXAI's recruiting team uses Grok Bot across the ATS, the coordinator routines in its public templates, and the candidate privacy rules they ship with.

Work value 3/10

Recruiting runs on many small, timely tasks spread across an ATS, calendars, transcripts, and Slack. SpaceXAI's recruiting team published how it uses Grok Bot across that stack, and two public templates by Tommy Hansen, listed under "From Grok Bot Team," cover coordination and sourcing. This playbook is for recruiters, recruiting coordinators, and hiring managers.

Not tested: The Daily Bot runs on a Bot computer without the Grok Bot desktop app, so we haven't connected an ATS or run these routines. Everything below comes from the linked guide, docs, and listings.

What recruiting uses it for#

Holly Dyche's recruiting guide describes four Bots the team built at an offsite:

  • Interviewer prep. "Right before the call, Preppy drops one Slack note that consolidates who you're talking to, why we're excited about them, interview stage, and what the focus areas should be." (Preppy install page)
  • Scorecards. Pulse "pulls the Granola transcript or meeting recording, matches it to the scorecard for that role, and drafts the interview feedback form in your voice," then asks for your take before it "redrafts and submits to your ATS." (Pulse install page)
  • Sourcing. Scout "sources candidates overnight, adds them to the ATS as leads, and drafts multi-touch sequences."
  • Loop analysis. eVal "looks across the loop for where candidates are failing and why."

Hansen's Recruiting Coordinator "schedules interview loops, preps your interviewers, and chases what's stalled." Hansen's Talent Discovery Bot "finds candidates for open roles that match your criteria and aren't already in your ATS."

Connectors to set up#

Cursor's talent scout starter says to "connect your applicant tracking system, approved sourcing tools, email, and calendar." Hansen's Talent Discovery listing integrates Slack, Notion, Linear, Figma, Gmail, Google Sheets, and Ashby (listing). Dyche's Bots move context between "Notion, Slack, and Ashby" (guide), and Pulse reads Granola transcripts.

If your ATS has no plugin, the Bot can use its own computer and browser; see our explainer on the Bot computer and our guide to sign-ins and secrets.

Routines#

The main routine, from the Recruiting Coordinator listing:

  • Daily hiring brief, "every weekday morning, today's interviews, who still needs scheduling, and what is waiting on someone."

Two more:

  • Afternoon loop check, "every weekday afternoon, what went stale since the morning brief, each item with a drafted nudge" (Recruiting Coordinator).
  • Daily candidate batch, "every weekday morning, sources a small batch of new candidates for the open role, deduped against your pipeline, with evidence and a source link on every card" (Talent Discovery).

Both templates start with routines off. "My routines stay off until you turn them on and they run in your timezone," the Coordinator says. Neither listing says what happens on a day with nothing stale or new, so add a line like "if nothing changed since the last brief, post one line or nothing." Our quiet routines guide explains the pattern, and Cursor's docs list what to confirm when you create one, including "what happens when a source is missing."

Example prompt#

The "Talent scout" starter from Cursor's use-case docs:

Prompt
For this role description, find 20 candidates who meet the must-have criteria. Exclude anyone already in our ATS, explain the evidence for each match, and draft personalized outreach in my voice. Don't contact anyone.

The docs add: "Keep external outreach behind approval, and have the Bot respect candidate privacy and each source's terms."

What to expect#

These outcomes are self-reported by the recruiting team, with no measured before and after.

  • Dyche: "Within a week we had pages of use cases, with recruiters sharing prompts and routines with each other" (guide).
  • On scorecards: "Scorecards stop being the thing you put off until 9pm."
  • Overall: "Most of my day is candidate-facing again: interviews, debriefs, and calibration, not scorecards and pipeline reports at 9pm."

Gotchas#

  • Candidate privacy is built in, keep it. The Coordinator will "never store or infer age, a graduation year used as an age proxy, gender, race, nationality, religion, disability, health, pregnancy, marital or family status, or sexual orientation" and will "never guess an email, a phone number, or an employer" (listing).
  • Nothing goes to candidates without you. "I never send, book, post, or reject without your yes" (Recruiting Coordinator). Scout's outreach follows the same rule: "You approve the send" (guide).
  • The tracker is the source of truth. "The tracker is the record, chat is not," per the Coordinator listing. Cursor's docs make the same point: "keep changing facts in the source system" (docs).
  • Teach it once, save a skill. Dyche: "walk it through your reasoning, then have it turn that into a reusable skill." See our tip on saving a process as a skill.
  • Routine approvals expire. Approvals from a routine "expire after about 10 minutes," per the FAQ.
  • Morning and afternoon checks add up. "Each run spends usage," per the routines help page.

Sources#