How to use an AI agent for job-search alerts that fit your next career move
Use an AI agent to check employer listings against your criteria, remember roles already seen, and send new matches or meaningful changes. Separate must-haves, preferences and unknowns before ranking anything. The result is a traceable shortlist for your review, not a prediction of who will hire you.
Define the move before the search
Write what would improve on your current situation: more ownership, different work, a pay threshold or a workable location. Supply a skills summary and example roles. A matching title alone does not establish that a move is worthwhile.
The open-source autopilot-jobhunt project documents nightly career-page searches, resume-based scoring and alerts, with applications left to the user. We have not run it. Our original adaptation checks gates first, preferences second and missing evidence throughout.
Use gates, preferences and an unknowns queue
Gates are conditions you will not trade away. Preferences help order roles that pass. Unknowns are facts the listing or your profile does not establish. Ask for a source reference beside every judgment. A resume that omits a skill does not prove you lack it; the agent should ask.
This fictional example concerns a worker seeking remote individual-contributor work. The criteria are theirs, not universal career advice. No role or salary below represents a current vacancy.
| Criterion | Type | Evidence | Decision |
|---|---|---|---|
| Remote from Colorado | Gate | Posting explicitly includes Colorado | Met |
| At least $140,000 base | Gate | No salary stated | Unknown; ask |
| More product ownership | Preference | Owns a roadmap; authority unclear | Partial; verify |
| Uses existing Python skills | Preference | Posting requires Python; profile includes a dated project | Supported |
| No direct reports | Gate | Posting requires managing six people | Unmet; exclude |
Keep the explanation stronger than the score
Use three queues: worth reviewing, needs clarification, and excluded by a confirmed gate. A preference cannot cancel a failed gate. Decide whether uncertain roles belong in a separate occasional digest. Missing salary should never silently become either a passing pay check or proof that the job pays too little.
If you want numbers, agree the preference weights first and show matched weight, assessed weight and unknown weight separately. Do not renormalize away missing evidence. The result describes your rubric, not your hiring probability. Lever’s official Postings API documentation identifies salaryRange as optional; a neatly formatted feed can still leave compensation unanswered.
Verify the employer source before including a role
Greenhouse’s Job Board API documents public read access to published jobs and distinguishes a job-post identifier from the underlying job identifier. Those fields can help maintain a watch; they do not make one feed a search of the whole market. Use a supported public feed or browser, and record inaccessible sources instead of bypassing restrictions.
- Start from the employer’s own careers page and follow its linked hiring-system page. Use aggregators for discovery, then inspect the current employer posting.
- Record the direct link, employer, posting ID when available, location restrictions, stated responsibilities, compensation terms and check time. Keep a missing posting date distinct from the date you found it.
- Read the full description before matching. Compare like compensation units and distinguish base pay from bonus or equity. Do not infer worldwide remote eligibility from a remote label.
Remember old roles without hiding real changes
Keep a private ledger keyed by employer and posting ID, falling back to a canonical employer URL when needed. Store first seen, last checked, last alerted, relevant field changes and your interested, skipped or applied decision. Group related location variants without assuming distinct postings are the same opening.
Ashby’s API defines publishedAt as when a job was last published. Treat a refreshed timestamp as something to investigate, not proof of a newly created role. Its posting documentation also distinguishes unlisted pages, still reachable by direct link, from unpublished postings. Absence from a board alone is not enough to announce a closure.
Preserve a role’s history when it returns. Label it relisted or changed with the evidence, and alert again only if it meets your chosen rule. A timeout belongs in the coverage log; it is not a successful check with zero matches.
Keep the resume private and the delivery explicit
Begin with a redacted skills summary. Omit contact details, references and confidential employer information the search does not need. Identify the profile version and approved storage location. Authorize uploads to named services separately; never put a private resume link in a public report.
Review one sample before scheduling. Choose a daily time and timezone, exact recipient and a maximum shortlist size. Prefer new matches and material changes only, with no filler on quiet days. Verify the saved schedule and delivery status. Without a supported trigger, “new-match alerts” means the next scheduled check, not instant discovery.
A starter brief to make your own
Watch [employers and sources] for [target work]. Use [redacted profile version]. My gates are [conditions]; preferences are [ordered list]. Check employer postings, keep source references, and separate met, unmet and unknown. Return at most [three] new roles, each with one reason it could improve my situation, one tradeoff and one question. Keep a dated ledger and preserve my decisions. Show a sample first. Then use [supported schedule, timezone and approved destination], notifying only on new matches or relevant changes. Do not upload my resume, contact anyone or apply without separate approval.
Check the misses as well as the matches
Open the original posting again before applying. You decide whether the work fits your goals and whether to proceed; the agent helps you organize the evidence. This tutorial and its sample brief are untested editorial adaptations.
- An obviously unsuitable role appears: inspect the exact gate and evidence; fix the rubric before increasing the search scope.
- A promising role is missing: check source coverage and exclusions. Revise your criteria deliberately rather than letting the agent relax them silently.
- An old role keeps returning: inspect its ID, previous decision and changed fields.
- Nothing arrives: check the last successful run and delivery record before concluding there were no matches.
Copy a prompt and try it
Each idea has a complete starter prompt, the inputs to bring, and the tools it needs.
- Get a daily email researched for your actual interests
A concise sourced email with only new matches
- Keep a decision journal that remembers what you knew then
Timestamped decision records and a follow-up review of assumptions and outcomes
- Stop writing the same proposal from scratch
A private proposal component library and a source-mapped sample draft
- Practice the negotiation before you are in it
A negotiation rehearsal, feedback and a one-page conversation card
Common questions
Can I start without uploading my resume?
Yes. Supply a redacted summary of relevant experience, examples and preferences. Keep uncertain qualifications as questions rather than letting the agent invent them.
Should a high fit score trigger an application?
No. Inspect the evidence, unresolved gates and current employer posting. A rubric score helps organize your own search; it is not hiring odds or permission to submit.
Will this work in Muse, Instinct, Dots or Grok Bot?
Use the brief with whichever assistant has the required research, persistent-record, scheduling and delivery tools. If a capability is missing, request a manual shortlist or email draft. This guide does not establish platform compatibility.
Sources and further reading
- tarunlnmiit: autopilot-jobhunt project documentation; design not independently run-tested — checked October 1, 2026
- Greenhouse: Job Board API, published jobs and identifiers — checked October 1, 2026
- Lever: Postings API, optional salary fields — checked October 1, 2026
- Ashby: Job Postings API, publication time and available fields — checked October 1, 2026
- Ashby: Job Postings, unlisted and unpublished behavior — checked October 1, 2026
These are practical editorial examples. The prompts have not been independently run-tested across every assistant.
Keep exploring
- Personal AI agent vs chatbot: which should you use?
- How to set up an AI morning briefing from email, calendar and notes
- AI personal assistant use cases: choose a job with a useful finish
- AI life admin prompts for the paperwork you keep putting off
- Recurring AI agent tasks that tell you when something matters
- Creative AI agent projects you can open, play and make your own
- How to adapt a personal AI agent prompt for Muse, Instinct, Dots or Grok Bot
For platform setup, read the Muse, Instinct, Dots and Grok Bot guides.