find your role first.

Guide

How do I automate my job search with AI in 2026?

To automate your job search with AI in 2026, automate the finding and the reading, and keep the applying for yourself; connect an agent like Claude Code or Codex to a job feed such as Find your role first, run it on a schedule, and let it hand you a short, reasoned list each morning. That removes most of the grind without sending a single application you didn't choose.

Auto apply is the part I'd leave out. Everything before it's fair game.

Which parts of a job search can AI automate?

More than people think, and less than the ads say.

StepAutomate?Why
Finding new rolesYesPure repetition, and a feed does it well
Filtering by title, place, dateYesMechanical
Reading postings against your resumeYes, with reviewAgents are good at this, not perfect
Ranking and writing fit notesYes, with reviewSaves the most time
Tailoring a resume or letterDraft onlyYou should still edit
Submitting applicationsNoJudgment, and your name is on it

Why not use an auto apply tool?

Auto apply tools exist, and to be fair, some people get interviews from them. The trouble is what they optimize for. Volume. A hundred applications with thin tailoring can look like exactly that to the person reading them, and a few bad fits sent under your name are hard to take back.

The slower version works better for most people I've watched do this. The agent does four hours of scrolling in a few minutes, and you spend twenty minutes applying to the three roles that actually fit. Find your role first is built around that split. It finds roles and returns the apply link. It doesn't apply.

What does the automated loop look like?

Four steps, run once or twice a day.

  1. Pull what changed. The agent calls get_changes with the cursor it saved last time, and gets back new, updated, reopened and closed roles since then.
  2. Filter. It keeps roles that match your titles and locations and drops anything closed.
  3. Read. For the ones that survive, it opens the original posting with its own web tool. The feed doesn't include job descriptions, so this is where the real reading happens.
  4. Report. It writes a short list to a file: title, company, why it fits, the biggest gap, the apply link. Then it saves the new cursor.

The cursor is what makes this cheap. Each run only looks at what's new. And on this feed, a role you've already seen that month is free to re-read and empty searches are free, so a quiet morning costs nothing.

How do I run it on a schedule?

Both agents can run a prompt from the command line without an interactive session. Claude Code has a print mode (claude -p) and Codex has codex exec. Put your instructions in a file and call one of those from cron, launchd or any scheduler you like.

A prompt file that works as a starting point:

Read cursor.txt. Call get_changes from that cursor.
Keep open roles whose title matches my targets in rules.md and whose location fits.
Open each kept posting and compare it with resume.md.
Append the best matches to shortlist.md with title, company,
two lines on fit, one line on the biggest gap, and the apply link.
Save the new cursor to cursor.txt. Don't apply to anything.

Then, for example, a morning cron line:

0 8 * * * cd ~/job-search && claude -p "$(cat daily.txt)"

Check your client's docs for how it handles tool permissions in non-interactive runs, since you'll want the job feed and web fetch allowed without a prompt.

What should I do with the shortlist?

Read it with coffee. Honestly, that's the whole point: the list is short enough to read properly.

Pick the roles you'd actually take. Ask the agent to draft a tailored resume line or a note to the team if you like, then edit it yourself and apply from the original apply link. Check get_job on anything that's been sitting a few days, since it'll tell you if the role closed.

How much of my allowance will this use?

Find your role first covers 3,000 new roles a month, up to 1,000 a day, for $9. A daily run that only pulls changes and filters them usually stays well inside that, and get_usage is free to call if you want the agent to check before it searches. Cancel anytime, refund within 3 days.

How do I set up automated job search with Find your role first?

Sign in with Google at findyourrolefirst.click/account, subscribe, create an agent key, and connect your client.

Claude Code:

claude mcp add --transport http --scope user jobfeed https://findyourrolefirst.click/mcp --header 'Authorization: Bearer YOUR_API_KEY'

Codex:

export JOBFEED_API_KEY=YOUR_API_KEY
codex mcp add jobfeed --url https://findyourrolefirst.click/mcp --bearer-token-env-var JOBFEED_API_KEY

Make a folder with resume.md, rules.md and your prompt file, run it once by hand to see the output, then schedule it. There's a REST API too if you'd rather write the loop yourself.

Questions

It can automate finding, filtering, reading and ranking roles. I'd keep applying manual, because that's where judgment and your name come in.

Does Find your role first apply to jobs automatically?

No. It returns the apply link and you apply yourself.

How does the agent know which roles are new?

It calls get_changes with a saved cursor, or searches with discovered_since. Both use the time the feed first saw each role.

How often should the automated search run?

Once or twice a day suits most people. Every board is checked every two hours, so running more often than that rarely finds much extra.

Do I need to know how to code?

Not much. You need to paste a setup command and, if you want a schedule, one cron line. The agent does the rest from plain English instructions.

Try it in your agent

$9 a month. Sign in with Google, subscribe, then paste one line into Claude Code or Codex. Cancel anytime, refund within 3 days.

Get started