Let your AI agent run the job search (you keep the send button)

If you already talk to an AI assistant every day, the most natural version of a job search is the one you never leave that chat for. Ask it to tailor your résumé for a role. Ask it what's moving this week. Tell it a recruiter replied. The repetitive parts of a search are exactly the parts agents are good at.

The catch has never been ability. It's memory. Sessions end, chats compact, models change. The context that mattered last week is gone, so you re-explain your history to every new conversation.

Give the agent somewhere to stand

The fix is boring and structural: your search needs to live somewhere outside the chat. A system of record the agent reads from and writes to (your applications, contacts, interviews, and what worked), so any session, in any tool, starts already knowing where things stand.

That's what MCP makes practical. It's an open protocol that Claude, ChatGPT, and a growing list of tools speak. Connect once, set the scope, and your agent can pull a job description apart, tailor against your real profile, file the application, and save the recruiter you mentioned in passing.

What to delegate first

• The weekly sweep: “check my email for updates on last week's applications” and let the statuses reconcile.
• Tailoring in batches. One ask, several roles, résumés ranked by fit.
• Contact capture. Say a name, a number, and a company once; never lose it.
• Status bookkeeping: “Eliot referred me” should be a sentence, not a chore.

What you should never delegate

The send. A good setup lets your agent draft the cover letter, the outreach message, the negotiation plan. It stops there. Applying and sending stay yours, from your own accounts, after you've actually looked. That split isn't a limitation; it's what makes the whole thing trustworthy. The agent takes the grind. You keep the judgment.