Built with Muse Publish a workflow
Task guide

Muse for getting a job

Finding the roles is the fun part. The job itself is an evening of portals, accounts, and forms that ask you to type your own resume back into boxes. People are now handing that transcription work to Muse, Meta's personal AI agent, and approving the results in batches each morning. This page shows the setup exactly as it has been documented, the prompts to copy, and what the reports do not yet prove.

Last checked: October 9, 2026 ยท 8 min read

The application treadmill

You find a role that fits, and the evening disappears. The company portal wants an account. The account wants your resume uploaded. The form then wants the same resume typed out again into boxes, one job at a time, plus a cover letter rewritten for a company you had not heard of that morning. Do that for five roles and you have spent the night on transcription, not on the part that might actually get you hired.

Look at the shape of that work. Reading job posts and judging fit against written criteria. Tailoring the same document again and again. Filling the same facts into similar forms, carefully, without inventing anything. That is tedious, rule bound work, and it is the shape of work an agent is built for. The documented setups below treat Muse exactly that way, as the operations team for your search, with you kept on approvals and judgment calls.

What people actually did

The fullest account is a workshop by Sid Saladi on Substack, Make Meta Muse Apply for Your Jobs. His setup has two parts: a skill that holds his resume, and a setup prompt that makes Muse write its own job search rules before it searches for anything. It then runs every morning and brings him a batch of applications to approve. One documented day: 17 roles found, 12 applications submitted, 1 company limit reached, and 4 roles parked for his input.

His hardest rule is about where applications go. Muse never applies through LinkedIn or Indeed. It reads their alert emails to discover roles, then applies on the company's own career site. The moments that stay human in his setup are just as deliberate: approvals, CAPTCHAs, sign in codes, passwords, and any question about salary or anything that needs an essay answer.

A second builder, Haytham Abouelfaid, described his version on LinkedIn: find roles, check fit against real criteria, prepare the documents, fill forms carefully, and keep a tracker with an audit trail. His first run was a discovery test on purpose: 10 roles reviewed and 0 applications submitted while he validated his sources and criteria. That zero is part of the record, and this page keeps it in.

The preparation side has its own documented run. A YouTube walkthrough by Agastya Intelligence, Muse Audited My Resume, has Muse read a resume the way an applicant tracking system would, then again the way a hiring manager would, and produce five fixes covering layout, weak verbs, the objective statement, the skills list, and keeping the resume to one page. The audit is written out in full in the workflow below.

The workflow, step by step

Step 1: Audit the resume before anything applies with it

An agent applying at volume with a weak resume just collects rejections faster. Start with that audit, written out here with the job description included so the fixes aim at a real target:

Read my resume twice. First as an applicant tracking system: show me what a parser would extract, what it would miss, and which keywords from the job description below are absent. Then as a hiring manager with ten seconds: the three things you notice first, the three things that make you doubt, and the exact lines you would rewrite. Never invent achievements. Work only with what is on the page. [paste your resume, then the job description]

Step 2: Teach Muse your resume once, as a skill

Save the audited resume as a skill Muse can load on every run, along with the facts that never change: your target roles, your locations, your salary floor, and the companies you will not work for. Every later run reads from this, so no run starts by guessing who you are.

Step 3: Run the setup prompt, and read the rules it writes

This is the heart of Saladi's setup: the setup prompt makes the agent write its own operating rules and show them to you first. Following his documented method, in full, it runs like this:

You are my job search agent. Before you search for anything, write the rules you will work by and show them to me: my target roles, my locations, my salary floor, the companies to avoid, and what counts as a strong match. My resume is the only source of facts about me, and you must never invent experience I do not have. Every morning, search for new roles, score the fit against your rules, draft the tailored documents, and bring me a batch. Submit nothing I have not approved. Apply only on company career sites. If a form asks about salary or needs an essay answer, park it and ask me.

Step 4: Approve the morning batch, then let it submit

Each morning the run ends in a batch, not in submissions. You see every role, the fit reasoning, and the parked questions, and you approve, edit, or drop each one. Only then does it apply, on the company's own site, logging every submission as it goes:

Run today's job search and build my approval batch. For each role: the company, the title, why it matches my rules, and anything you are unsure about. I will approve, edit, or drop each one. Submit only what I approved, on the company's own site, and log every submission with the date, the role, and the link.

Step 5: Keep the human moments human

CAPTCHAs, sign in codes, and passwords are yours. Salary questions and essay answers are yours. Anything ambiguous gets parked for you rather than guessed at. That split is not a limitation of the setup. In the documented versions, it is the setup.

Where it falls short

The record so far documents effort, not outcomes. Saladi's log shows roles found and applications submitted. Abouelfaid's shows roles reviewed. Nobody in the published record has yet reported the numbers that matter most, interviews and offers, so judge this workflow as evenings saved and applications actually sent, not as a promise about hiring. A search that applies badly at scale is still a bad search.

Company limits are real: one was hit on the first documented day, which is part of why four roles were parked instead of forced. Portals also defend themselves, which is the practical reason the documented setup routes around the big job boards and keeps the human on codes and CAPTCHAs. And the validation zero matters. A setup that has not had its sources and criteria checked will happily spend your mornings on the wrong roles. Run a discovery test first, like Abouelfaid did, and read the batch closely for the first week.

One caution. Your resume and your application history are personal documents. Before you load them, turn off the data sharing setting under Settings, then Data controls, so your chats are not used to improve the data model. Our guide to whether Muse is safe covers the settings and permissions worth reviewing before you connect anything personal.

Job hunting workflows on Built with Muse

Our earlier guide, Muse for job hunting, covers the manual version of this search, prompt by prompt. The workflows people publish here cover the surrounding jobs: getting the useful bits from ten daily newsletters is the same triage habit pointed at your inbox, and daily inbox triage keeps the alert emails this setup depends on from burying you.

Nobody has published a job application workflow here yet. If you set this up and run it, publish your version: your rules, your prompts, your numbers. The first one will be the page every later applicant copies, and it will carry your name.

Get started

Get a code

Set it up tonight, approve a batch tomorrow. The setup on this page takes one sitting: an audit, a skill, one setup prompt. After that the work moves to a short morning approval. Get a free code from the community pool to start, or join the waitlist if Muse has not launched in your country yet.

Get a code Join the waitlist

Frequently asked questions

Will Muse submit applications without asking me?
Not in the documented setup. The agent searches, scores, and drafts, then brings you a batch. Submissions happen only for the roles you approved. If a version you are following applies without an approval step, change that first. It is the rule that keeps a bad match from going out under your name.
Why does the setup avoid LinkedIn and Indeed?
By design, in Saladi's workshop. The big boards fight automation and bury applications in volume. His setup reads their alert emails to discover roles, then applies on each company's own career site, where the application lands in the employer's own system.
What do I still have to do myself?
Approvals, CAPTCHAs, sign in codes, passwords, salary questions, and essay answers. The agent also parks anything ambiguous instead of guessing. Expect the human share to shrink to a short morning review once the rules are tuned, but plan to read the first batches closely.
Will this get me interviews?
The published reports do not say yet. They document roles found and applications submitted, 17 and 12 in a single documented day, and they stop there. Treat the workflow as a way to run a serious search without spending your evenings on forms, and judge it on your own reply rate after a few weeks.
Do I need an invite code to run this?
Yes. Access to Muse runs on invite codes today. You can claim a free code from the community pool on this site and do the setup the same day. If Muse has not launched in your country yet, join the waitlist and you will get a code when it does.

Built with Muse is an independent community project. Not affiliated with or endorsed by Meta.