Can an AI agent run your whole job search?

Partly, and the part it can't do is the point. An AI agent can scan job boards, evaluate each listing against your CV, tailor a résumé, and track every application. What it should not do is press apply for you: mass auto-apply burns your reputation and misfires. career-ops runs the whole pipeline as a human-in-the-loop system; the agent does the work, you make the call.

What an AI agent can automate today

A capable agent already handles the mechanical weight of a search, end to end:

Scan. Pull open roles from company career pages and ATS APIs (Greenhouse, Ashby, Lever) without you opening thirty tabs.
Evaluate. Read each job description, score it against your background across five dimensions plus a holistic global score, and cite the specific lines of your CV and the JD that drove the number.
Tailor. Generate a résumé tuned to the role in front of you: the same CV, re-emphasised, never invented.
Track. Log every application to a local file so you have one source of truth instead of a spreadsheet you stopped updating in week two.

That is four stages of real work the agent does while you sleep. The fifth stage, pressing apply, is the one it deliberately hands back to you.

Why mass auto-apply fails

The loud promise of the last two years has been the opposite: an agent that fires hundreds of applications a day on your behalf. It reads well and works badly.

Applying is a one-shot, reputation-bearing move. Send a generic application to a role you half-fit and you don't just miss that job, you teach a recruiter, and sometimes an entire company's ATS, to down-rank your name. Volume is not leverage here; precision is. An agent that optimises for applications-sent is optimising the wrong number.

So the honest answer to "can an AI apply to jobs for me?" is: it can prepare every application to the point of one click (the evaluation, the tailored résumé, and the tracker row are all done) and then it stops. You decide what to send, and you send it. That is not a limitation bolted on; it is the design. The agent removes the tedium, not the judgment.

Running it with Claude Code, or any AI CLI you already pay for

career-ops does not ship its own model. It is a set of prompt files and slash commands that run inside an AI coding CLI you already use, and the reasoning comes from whichever model you point it at.

Claude Code is the most common way people run it, because its skill loader makes the modes one command away, but it is a flagship path, not a requirement. The same files run on Codex, OpenCode, Qwen, GitHub Copilot CLI, and others, and you can bring your own model or a local one. Pick the engine that fits your cost, quality, and privacy profile; career-ops runs on top of it unchanged. Naming Claude Code here is descriptive, not an endorsement; you are never locked to one vendor.

Everything runs on your own machine. Your CV, your profile, and your application history never leave it unless you push them somewhere yourself. career-ops keeps zero telemetry: there is no product-side dashboard of what anyone scanned or applied to.

What one real search looked like

Because there is no telemetry, the only numbers we can honestly show are personal ones. In my own 2026 job search, career-ops evaluated 740 listings; I applied to 68, got 12 interviews, and 1 offer.

The interesting brick is the middle: 68 applications turned into 12 interviews, a 17.6% interview rate. That is what "the agent does the work, you make the call" buys you: not more applications, but fewer, better-chosen ones, each prepared well enough that a much larger share of them turned into a conversation.

Limitations: where an agent still can't help

An honest page names the edges. career-ops is a strong pipeline, not magic:

It doesn't know the room. The agent researches a company from public sources; it cannot tell you the unwritten dynamics of a specific team. That is still your network's job.
Scoring is a rubric, not an oracle. The first few evaluations are rough: the system does not know you well yet, and you correct it as you go. A 4.7 is a strong signal, not a guarantee.
Liveness is heuristic. It checks whether a listing is still open, but a job marked live can still be filled internally, and a closed one can still be worth a referral.
It won't negotiate for you. It preps you for the comp conversation; you have it.
It is for an active, structured search. If you are sending one or two applications, the fifteen-minute setup is not worth it. This is a tool for people running a real search, comfortable in a terminal even if they don't code professionally.

The part an AI agent can't do (read a room, hold a conversation, decide what you actually want) is the part that was always yours. career-ops just clears everything else off your desk so you have the attention for it.

Frequently asked

Can an AI agent apply to jobs for me? It can prepare every application to one click: the evaluation, the tailored résumé, and the tracker entry are all done automatically, but it does not submit for you. You review and send each one yourself. That is deliberate: mass auto-apply damages your standing with recruiters and ATS systems, so career-ops removes the busywork and keeps the decision with you.

Is it safe to let an AI agent handle my job search? career-ops runs entirely on your own machine, inside the AI CLI you already use, and keeps zero telemetry: your CV, profile, and application history never leave your computer unless you push them somewhere. It reads and prepares; it never submits, deletes, or shares anything without your explicit action.

Do I need Claude, or does any AI CLI work? Any supported CLI works. Claude Code is the most common runtime because of its skill loader, but career-ops is engine-agnostic: the same prompt files run on Codex, OpenCode, Qwen, GitHub Copilot CLI, and more, and you can point it at your own or a local model. Several options run at $0.

Ready to try it? Start with the Quick Start or read what career-ops is.