Automating the Browser: Job Applications at Scale

Applying to jobs is the worst kind of repetitive work. Same fields, same resume upload, same “Tell us about yourself” textarea — 50 times over. This is exactly what agents should handle.

The Browser Automation Stack

There are two approaches to browser automation with AI agents:

Approach	Tool	Best For
Headless scripting	Playwright / Puppeteer	Predictable, structured pages
AI-driven navigation	OpenClaw browser tool	Dynamic forms, varied layouts

The key insight: ATS platforms (Greenhouse, Lever, Ashby, Workday) are structured enough for AI-driven automation but varied enough that hard-coded scripts break constantly. AI agents adapt to layout changes that would crash a brittle script.

How OpenClaw’s Browser Tool Works

OpenClaw gives agents a browser they can control through natural language actions:

// Agent sees a snapshot of the page DOM
browser({ action: "snapshot" })

// Returns accessibility tree:
// [button "Submit Application"]
// [textbox "First Name"]
// [textbox "Last Name"]
// [file "Resume Upload"]

// Agent fills fields by reference
browser({
  action: "act",
  request: { kind: "fill", ref: "First Name", text: "JD Davenport" }
})

The agent sees the page as an accessibility tree — not pixels, not raw HTML. It understands form structure, button labels, and navigation elements. This means it works across different ATS platforms without platform-specific code.

Example: Greenhouse Application

Greenhouse is the most common ATS. Here’s what the agent does:

Step 1: Navigate to the Job Posting

browser({
  action: "navigate",
  targetUrl: "https://boards.greenhouse.io/company/jobs/12345"
})

Step 2: Snapshot and Analyze

The agent takes a snapshot and identifies all form fields. Greenhouse forms typically include:

• Personal info (name, email, phone)
• Resume upload
• Cover letter (optional)
• Custom questions
• EEOC/demographic fields

Step 3: Fill Standard Fields

// The agent maps its knowledge to form fields
browser({ action: "act", request: {
  kind: "fill", ref: "first_name", text: "JD"
}})
browser({ action: "act", request: {
  kind: "fill", ref: "last_name", text: "Davenport"
}})
browser({ action: "act", request: {
  kind: "fill", ref: "email", text: "jd@example.com"
}})

Step 4: Upload Resume

browser({
  action: "upload",
  paths: ["/path/to/tailored-resume.pdf"]
})

Tip

Always tailor the resume to the specific job before uploading. A resume generator that customizes bullets and keywords for each role dramatically improves response rates.

Step 5: Handle Custom Questions

This is where AI shines. Custom questions vary by company:

• “Why do you want to work here?”
• “Describe a time you led a team.”
• “What’s your expected salary?”

The agent uses context from its memory (your background, preferences, salary expectations) to generate appropriate answers. No copy-paste from a template — each answer is contextual.

Step 6: Review and Submit

// Take a final snapshot to verify all fields
browser({ action: "snapshot" })

// Agent reviews: all required fields filled? Resume uploaded?
// Then submits
browser({ action: "act", request: {
  kind: "click", ref: "Submit Application"
}})

Handling Edge Cases

CAPTCHA

Some ATS platforms use reCAPTCHA. Options:

• 2Captcha service — sends the challenge to human solvers (~$3 per 1000)
• Wait and retry — some CAPTCHAs only trigger on suspicious behavior
• Skip and flag — mark the application for manual completion

Caution

Respect rate limits. Submitting 100 applications per hour will trigger anti-bot defenses. Space applications 3-5 minutes apart and use realistic browsing patterns.

Multi-Page Forms

Workday loves multi-page applications. The agent handles this by:

1. Completing each page
2. Clicking “Next” or “Continue”
3. Snapshotting the new page
4. Filling the next set of fields
5. Repeating until submission

Login-Required Applications

Some companies require account creation. The agent can:

• Create accounts with a dedicated email
• Store credentials in memory
• Log in for future applications to the same company

The Ethics Question

Automated job applications get mixed reactions. My take:

1. You’re not gaming the system — you’re eliminating busywork
2. Every answer is still yours — the agent uses your real experience
3. Quality matters — a tailored resume and thoughtful answers beat mass-spray
4. Time is finite — spending 45 minutes per application on 50 applications is 37 hours of repetitive form-filling

The agent doesn’t fake credentials or fabricate experience. It translates your qualifications into their form fields. That’s automation, not deception.

Getting Started

1. Set up your first OpenClaw agent
2. Create your base resume and candidate profile
3. Build a job search agent that finds relevant postings
4. Connect the browser automation for applications
5. Track results in a spreadsheet or database

The whole pipeline — search, tailor, apply, track — runs while you sleep.

---

About the author: JD Davenport builds AI agent systems at OpenClaw. Follow on LinkedIn for updates on building AI agents for business.

Enjoyed this? Get weekly agent architecture insights.

Subscribe and get a free Agent Architecture Cheatsheet.

Subscribe