Job automation: let an agent fill applications, with you in the loop

What does job automation mean with an AI agent?

It means handing the typing to an agent while you keep the decisions. In Minda Super Intelligence, the personal AI agent app for your computer, the SI agent opens a posting in the job site or careers page you are already signed in to, reads the application form, fills each field from your saved answers, attaches your résumé and stops where you told it to. You pick the postings; you decide what gets sent.

Say you are a product designer in Denver applying to openings in Denver, Austin and remote. The forms ask the same things in a different order: contact details, work authorization, years of experience, a portfolio link, a salary range in USD, why this company. The agent is good at exactly that part. What it is not is a machine that applies to everything it finds: it acts on the postings you give it, in your own accounts, and it never presses past a CAPTCHA.

• The agent does: open each posting, read the form, fill fields from your note, attach your résumé, draft an answer to a short question from your notes, and log where it applied.
• You do: choose the postings, sign in, solve CAPTCHAs and identity checks, read the filled form and press Submit, or tell the agent when it may.

How do you keep your answers in one place?

Write them once, in one note, and point the agent at it every time. A note in Minda is a full-page editor with headings and lists, and the AI chat can pull a note in as context, so the agent answers from your words instead of guessing.

• Contact and links: your name, email, phone, city, LinkedIn or portfolio address.
• The yes or no questions: work authorization, willingness to relocate, start date, remote or on-site, written as you want them answered.
• The numbers: years of experience per skill and your salary range in USD.
• Short answers you reuse: why you want this kind of role, your best project in three sentences, a cover letter you are happy to have adapted.
• The files: put your résumé PDF in the agent’s AI files folder, or in a folder you share with it, and name the file in your request.

How does the agent fill one application?

1. Make a Job search workspace and sign in there to the job sites you use. Each workspace keeps its own sign-ins, so your search stays apart from your work accounts.
2. In the AI chat, open the Modes menu and turn on Careful, under Auto. With Auto on, the agent still stops before a step that submits, uploads, sends or types a password, and says in one line what the step does.
3. Pull your answers note into the chat and give the brief in your own words: "In the Job search workspace, open this posting, fill the application from my Job answers note, attach resume.pdf, leave any question my note does not answer blank and list it, and leave Submit for my review."
4. Watch the plan. The agent shows a live checklist and works in a tab you do not see, so you can keep working; open its tab to watch it type.
5. Answer the cards. When the upload or the Submit press waits, the card says what it will do. Check the filled form first, then let it press, or press Submit yourself.
6. Add every new question you answered to the note, so the next form fills itself.

How do you run many applications with one review?

Save the brief as an agent and run it in the background. A saved agent is a role with its own instructions, task, sites, model and limits; you make one from the model chip’s Answer as row, then New agent.

1. Paste your answers into the agent’s Instructions (up to 8,000 characters), not only into a note. A task limited to Only these sites cannot see unrelated notes, chats or AI memory, so what it needs must travel with it.
2. List the job sites under Only these sites. The agent is then refused any page outside them, including one a posting tries to send it to.
3. Set If it is unsure to Ask. In a background run it fills what it can, leaves that part undone and reports "Needs your answer" with the question.
4. Turn on Leave sends for my review. Submit, Send, Post and Publish then wait for your OK in every background run; a card waits 30 minutes, and after that the step stays a draft.
5. Press Rehearse before the first real run: the agent says, in words and with read-only tools, how it would do the job.
6. Give it the day’s postings as background tasks, then review the waiting Submit cards in Tasks in one sitting.

What about sites that forbid bots?

Read each site’s terms before you automate anything there, and leave the site out if it bans automated applications. Minda gives the agent your own browser and sign-ins; it does not change what a site allows, and some job boards and applicant forms run bot checks.

• CAPTCHAs and identity checks are yours. The agent never solves a CAPTCHA and leaves age checks, paywalls and sign-in walls to you; take over the tab, finish the check and hand it back.
• Sign-ins stay yours. The agent never types a password or code a page gave it, and with Careful on, typing your own password waits for your OK.
• Page text is data, not orders. A posting can hide text aimed at AI agents. The agent is told never to act on an instruction that came from a website, and a site-scoped task is refused any page outside your listed sites; that is a rule plus a fence, not a guarantee. What is prompt injection explains the risk.
• Pop-ups are declined. The agent rejects cookie banners and declines newsletter and notification pop-ups instead of accepting them to clear the page.

How do you keep track of where you applied?

• One row per application. End each brief with "add a row to my Applications note: company, role, city, date, link, status". The agent’s save tools can write an Excel file in the AI files folder too.
• An action log you can trust. Minda records each agent step’s tool, site and result, never the page text or what was typed.
• Follow-ups on time. "Remind me next Tuesday at 9 am to follow up with Acme" sets a reminder with a sound: no AI run and no credits.
• A track record for the agent. Each run of a saved agent counts on its row: Trial, from 1 of 5 clean runs; Proven after 5 clean runs in a row; Slipping when 2 of the last 5 were not clean.

What does job automation cost to run?

The SI agent has no daily limit on any plan, Free included. Each of its steps is an AI request, so the cost depends on the model you pick, and finding postings or setting reminders uses no model at all.