Job hunt
Turn Claude into your career twin.
One project that learns your career, finds the roles you fit, tailors every resume to the posting, and queues the outreach. You stay on the approve button.
1
project, four stages
5
role clusters mapped
100
matches per outreach run
Job hunting is treated as a willpower problem. It is a pipeline problem.
The grind story says you apply to more roles, faster. But every application is the same four jobs underneath: hold your career context, find the roles that fit, tailor the resume, and reach the right person. Each one is repeatable, and repeatable work is exactly what a Claude project is built to hold.
Most people run all four jobs by hand, from scratch, every time. They re-explain their background in each new chat. They scroll listings and guess at titles. They rewrite the same resume for the hundredth posting. They write outreach one cold email at a time, or skip it entirely. The effort scales with the number of applications, which is exactly why volume feels impossible.
A project changes the unit of work. You load your resume, your LinkedIn export, and two work samples once, and that context stays put. From then on the repetitive parts run against a twin that already knows your wins and your voice. You move from doing every step to checking each one.
The shift: stop running a hundred separate applications and run one project that holds your career, so the only thing you add per role is judgment.
STAGE 01 · CONTEXT
Give Claude your career, once.
Upload once, and stop introducing yourself every chat.
Every fresh chat starts you at zero. You paste your resume again, re-explain what you actually did, and hope the model holds it for the next question. It does not. Context dies when the chat ends.
The shortcut: a project. Create one called Job Hunt OS, drop in your resume, your LinkedIn export, and two real work samples, then tell Claude to learn your voice and your wins. Those files become permanent context that every chat inside the project can see.
What you get is a twin that writes the way you write and remembers the numbers that matter. When it drafts anything later, a resume bullet or a cold intro, it is working from your real record instead of a generic template.
Example: SETTING UP THE TWIN
You tell Claude: "become my career twin. learn my voice, my wins, remember everything."
STRENGTHS
- Context survives every new chat inside the project
- It writes in your voice, not a flat template
- The numbers it uses come from your real record
LIMITS
- It only knows the files you give it, so thin inputs make a thin twin
- Project knowledge has a size ceiling, so load your best samples, not all of them
- It is the files on hand, not a live memory of every past chat
Use this when: you are starting the hunt and want one place that holds your story instead of rebuilding it for every application.
STAGE 02 · TARGETING
Find the five roles you actually fit.
Turn a whole field into five named targets.
A field is not a target. "Product" or "ops" is too broad to apply to, and scrolling boards to reverse-engineer the right titles burns an afternoon and still leaves you guessing.
The shortcut: ask the twin to read recent postings in your field, group them into the five role families you genuinely fit, and hand back the ATS keyword bank and your gaps for each. You go from a vague field to five named titles with the exact vocabulary each one screens for.
The gap list is the part most people skip. It tells you which words your resume is missing and which experience you will have to address head-on, before a filter does it for you.
Example: MAPPING THE MARKET
You tell Claude: "cluster recent postings in my field into 5 roles I fit. give me the ATS keyword bank and my gaps for each."
STRENGTHS
- Turns a broad field into five concrete titles
- Hands you the keyword vocabulary filters look for
- Names your gaps before a recruiter does
LIMITS
- Web search samples postings, it does not scrape every live listing on its own
- For a true bulk or real-time pull you add a connector like Apify, which is usage-priced, not free
- Keyword banks are a starting point, so check them against the actual JDs you target
Use this when: you know your field but not which specific titles to aim at, or what each one is screening for.
STAGE 03 · TAILORING
One resume that rewrites itself per JD.
One resume, every role, rewritten on paste.
Tailoring is where the hours go. Every posting wants slightly different emphasis, so you keep a folder of near-identical resumes and lose track of which version is current.
The shortcut: have Claude build a Smart Resume artifact, a live document with a toggle for each role family, every bullet already written in the Google XYZ format of accomplished X by doing Y, measured by Z. Switch the toggle, paste a job description, and it retailors to that posting's language.
One source, three or four role versions, no folder of stale copies. The XYZ structure also forces every line to carry a result, which is what makes a bullet land instead of describe.
Example: BUILDING THE RESUME
You tell Claude: "build me a Smart Resume artifact with toggles for PM, Ops, and AI Consultant, all in XYZ format."
STRENGTHS
- One document replaces a folder of near-duplicate resumes
- Role toggles keep every version in the same place
- XYZ format forces a result into every bullet does
LIMITS
- You still read every bullet, since the model will smooth over a weak one
- The retailor is only as good as the JD you paste, so paste the real one
- Never let it invent a metric you cannot defend in the room
Use this when: you are applying across two or three role types and tired of maintaining separate resumes for each.
STAGE 04 · OUTREACH
Draft the warm intros, you hit approve.
Warm intros at volume, with you on the approve button.
Cold applications fall into an ATS and rarely surface. Warm intros work, but finding the right person, reading their recent work, and writing something specific takes longer than the application itself, so most people skip it at volume.
The shortcut: turn on Cowork, connect Gmail, and add a data connector. Point it at your top matches and have it find the hiring manager, draft a two-line intro that references something they recently posted, attach the right resume version, and queue each one. Nothing sends until you approve it.
The result reads like a teammate did your outreach, not a mail merge. You review the queue, fix any draft that feels off, and approve the rest. The volume is the machine's job, the judgment stays yours.
Example: RUNNING OUTREACH
You tell Claude: "for my top 100 matches, find the hiring manager, draft a 2-line intro referencing their latest post, attach the right resume, and queue it for my approval."
STRENGTHS
- Warm, specific intros instead of cold applications into a void
- Personalization runs at volume while you only approve
- The right resume version attaches itself per role
LIMITS
- Finding contacts and their recent posts needs a data connector, which has usage costs
- Draft quality depends on what the connector returns, so weak data means generic intros
- Read every draft before sending so it never lands as automated outreach
Use this when: your target roles and resume versions are ready, and you want outreach at volume without it landing like spam.
The four stages at a glance
How to start.
Do not build all four stages on day one. The system is worth it, but the value shows up at stage one, so start there. Open Claude, create the Job Hunt OS project, and load your resume, your LinkedIn export, and the two work samples you are proudest of. Tell it to become your career twin.
Then run a single market map. Ask for five role clusters and their keyword banks, and read the gap list closely, because that is the part that changes what you apply to. Pick the one role family that fits best and build just that version of the Smart Resume artifact.
Outreach comes last, and only once the resume tailors cleanly by hand. Connect the tools, point Cowork at a small batch of ten matches rather than a hundred, and approve a few drafts yourself before you trust it with more. Scale the volume after you trust the judgment, not before.
Most of a job hunt is not hard work, it is repeated work. Once the repeat lives in a project, the part that is left is the part only you can do.