How Do I Use an AI Agent in My Job Search Without Letting It Contact Anyone?
Three things from my session, set in a job seeker's day: research from public pages, a contact tracker practiced on fake data, and replies that stay drafts.
Keep the tools that can reach other people on read only, practice with made-up contacts before you use real ones, and keep every reply a draft that you send yourself. It can research and organize. The sending stays with you.
A job search puts a lot of other people's names in front of you at once: recruiters, hiring managers, old colleagues you're thinking of writing to. An AI agent can help you keep track of them and get ready for the conversations. It shouldn't write to any of them on its own.
I didn't build a job search AI agent in the session. I built a chief of staff, and three things from that build carry straight over to a job seeker's day.
Can it research the people I'm about to meet?
It can, from public pages. I use my chief of staff AI agent for meeting prep, and I said I have it look at whether an organization has been in the press recently, shared a press release, or is celebrating something (30:19). For an interview, that organization is the employer.
The file on the session page keeps the research tools to public information:
From the file on the session page. Your research connectors read public information only. None of them log into a private account, and none of them post anything anywhere.
For the interviewer or someone you're writing to, the same rule means their public page and nothing behind a login. Can an AI Agent Read LinkedIn Profiles? covers what I said about that.
Can it keep a contact list without writing to anyone?
It can, and I'd practice on made-up people first. I told the group I have my AI agents create spreadsheets too, and this is how I asked for one when I showed my systems:
From the recording, 54:37. Let's see with demo names for a test with fake emails and phone numbers.
If you want to try email, I said to start with a dummy account, the kind most of us keep for promotional email (20:06).
Example, replace with yours: a sheet with one row per person and columns for company, role, how you know them, the date you last wrote, and what comes next. Fill the first ten rows with invented names. Ask the AI agent to tell you who is overdue for a follow-up. Once that reads right, swap in the real ones.
Can it draft replies I send myself?
It can. A recruiter's email, a thank-you note and a reply about an interview time are all things to read before they go. The file on the session page lists what the AI agent never does:
From the file on the session page. You never accept, decline, or respond to a calendar invite on my behalf. You never post, publish, or message anyone outside this chat that I haven't explicitly told you to reach in that turn.
An interview invite is a calendar invite. That line covers it. A written instruction asks. The tool settings are where you take the access away: on the tool management screen I can restrict or give permission, with read only tools in one section and write and delete tools in the next (43:59 to 44:14).
What Are Read-Only Tools for an AI Agent? explains the screen, and How Do I Control What an AI Agent Can Do in Gmail? goes through the email side, and What Does Human in the Loop Mean for an AI Agent? says what the documentation lists as the default setting.
What goes wrong
A tool with more access than the job needs. If you connect your main inbox on day one, everything in it is within reach. I suggested a throwaway account first, and I said the extra step is a good way to get comfortable with the level of access you're giving a system (19:58).
Real people's details on a screen others can see. I used demo names when I showed my systems because I didn't want to share personal information from people I work with (55:04).
A message that goes out unread. The file tells the AI agent to keep replies as drafts and not to send or post unless you say so in that turn.
What the recording doesn't show: a job search, an application, or a result from either. I'm not claiming one here.
Where this comes from
This article grows from the session I taught live with Women Applying AI on October 7, 2026. The quoted passage from the recording is my own words, taken verbatim, and the other lines from the recording are restated with their timestamps. The passages marked as from the file on the session page come from the files published with that session. The steps, tables and checklists around them were drafted by my team's AI agents from those words and my published pages, and checked against the recording before publishing.
Build Your First AI Agent in One Hour: the full session with the clips, the prompts and the transcript. The files are on the session page.
Related reading: How Do I Control What an AI Agent Can Do in Gmail?, Which Connectors Does an AI Agent Need?, and What Should Your First AI Agent Be?.
Frequently asked questions
Can an AI agent help with a job search if it can't contact anyone?
The three things from my session carry over: research from public pages, a contact list and drafted replies. The sending stays with you.
How do I practice before I use real contacts?
Use made-up names with fake emails and phone numbers, as I do when I show my systems, and a throwaway email account if you want to try email.
Where do I stop it from sending email?
In the tool settings, where you can restrict write and delete tools. The instructions file on the session page also keeps every reply a draft.
Build something that actually runs your workflow.
A focused, free 60-minute live session with Jacqueline. You build alongside her, on your own real task, and leave with an AI agent that is already running.