Most job searches fail on arithmetic. You are one person applying to roles one at a time, while the system on the other end is filtering on keywords your CV may not contain. The volume problem and the matching problem compound, which is why a strong candidate can send forty applications and hear nothing back. The work is not hard, there is just far too much of it for one person to do by hand, and so it gets done badly or not at all.
This workflow treats both problems as one job. It is four prompts, and only the third one needs agent mode. The insight worth holding onto is that a job search is not really a writing task. It is a matching task with a writing task stapled to the end, and Claude is unusually good at the matching half because it can hold keyword rules and a ranked list in view at the same time you are looking over its shoulder.
Step one: let Claude tell you what you actually qualify for
Upload your CV and ask it to act as a senior recruiter. The useful version of the prompt asks for two things at once: the twenty job titles you are genuinely qualified for, and the exact keywords each ATS scans for. That second half is what most people skip, and it is the half that decides whether a human ever sees the application. An applicant tracking system does not read for quality. It reads for presence, and it reads against a specific posting, so a role you are perfect for can be invisible to a system that never sees the word it wanted.
Asking for the titles first, before any writing, changes the shape of the whole search. You stop guessing at what to apply for and start from a list the model built off what you already have. That is a bounded, inspectable starting point, and you can argue with it before a single application goes out.
Step two: one master CV, not forty versions
Ask Claude to rewrite your CV into a master template you can tailor instantly. One structured source document, with the step-one keywords in the right places. Every tailored version after this is an edit, not a rewrite. This is the difference between a sustainable workflow and a grind: forty separate rewrites will burn you out by the tenth application, while forty edits to one good source is a few minutes each.
Keep the keywords visible in the master file. When a posting wants a phrase you do not have, you add it once and it propagates, rather than being rediscovered on every application. The master CV becomes a living document that gets sharper as the search proceeds, which means the hundredth application is better than the first, not wearier.
Step three: let it browse
Turn on agent mode and give it the search. Go to LinkedIn and Indeed, find every job matching those titles, and build a spreadsheet with a match score and a tailored CV for each. The spreadsheet is the deliverable that matters, because it turns an unbounded search into a ranked list you can inspect before anything goes out. You are not asking for applications yet. You are asking for a map of the territory, with a number attached to each point on it.
Step four: apply, with a tailored CV each time
Then the instruction that saves the hours: apply to the top hundred matched jobs, customising each application to the job description. Claude works the list while you do something else. Because the matching already happened in step three, this step is mechanical rather than creative. Each application draws on a CV that was tailored back when the shortlist was built, so there is no scramble at send time.
Where it breaks
Three honest limits. Agent mode browsing LinkedIn is the fragile part; the site does not love automation, so supervise the search step the first time and expect it to need a nudge. A match score is a model judgement, not a decision, so read the top of the list before you commit to it. And tailoring clears the filter, it does not make you qualified, so do not let the volume outrun your own judgement about fit.
There is also a subtler failure mode worth naming: volume can feel like progress. A hundred applications going out looks like motion, but if none of them were matched to a role you can actually win, you have industrialised rejection. The match score exists precisely to stop that, which is why reading the shortlist is not optional.
Used as a filter, this is a force multiplier. Used as a volume machine with no human read of the shortlist, it just gets you to a hundred rejections faster.




