Build an Agent That Runs Your Job Hunt End to End

Most people use an AI agent like a faster search box, one question at a time, then do the actual work themselves. Rohit hands his a job he does not want to do again, and walks away from it.

Rohit Reddy, AI Product Lead at Meta, starts from an empty prompt. Live, he gives the agent a resume and a set of target roles, watches it match openings and draft a tailored resume and cover note for each one, sets the approval gate that stops it before anything gets submitted, and wires it to his inbox so it comes back to him when a reply lands. The same prompt builds a dashboard that tracks every application, its status, and which recruiters have replied. One prompt, one agent, a workflow that is still running after the session ends.

You leave with the prompt and the pattern, ready to point at a task of your own.

What you'll walk away with

1. Scope an agent around a workflow, not a prompt. The difference between output and ownership is a trigger, a cadence, and a memory of what it already did. Rohit sets all three live, and you see exactly where each one sits inside the prompt.
2. Put the approval gate where the risk actually is. Most human in the loop designs check everything or nothing. This one drafts and matches freely and stops at the single irreversible step, which is the only place review buys you anything.
3. Make the agent report on itself. The same prompt that does the work builds the dashboard that tracks it and nudges him when a reply arrives. An agent that cannot show you its own state, or get your attention when something changes, has just handed you one more thing to check by hand.