Agentic job search automation

Turning unemployment anxiety into a system that works while I sleep.

Renee Romero • AI Systems Designer • 2026

A case study in multi-agent decomposition, context architecture, and evaluation design.

What this proves about how I design

Multi-Agent Decomposition

Broke one complex problem into 9 specialized, coordinated workflows that pass data seamlessly across stages.

Failure Pattern Recognition

Diagnosed and resolved context degradation, API overload patterns, specification drift, and cascading failures across 32 hours of debugging.

Context Architecture

Built 6 interconnected Notion databases that supply agents with the right information at the right time.

Cost & Token Economics

Runs at ~$6 per 40-page Company Intelligence Brief, ~$70/month total system cost.

Evaluation Design

Built self-scoring evaluation loops into resume and cover letter generation that rate, critique, and improve drafts before delivery.

Trust & Security Design

Human oversight by design. The system generates drafts, never sends automatically. Final decisions stay human.

SECTION 01

The problem

What job searching actually felt like.

Job searching is broken

5

job alert sources cluttering my inbox

90 min

to tailor each resume manually

~10

companies I could realistically track

I'd wake up knowing I should check job boards, but I'd avoid it. When I did look, it was chaos. And the worst part? That constant low-grade anxiety of "I should be doing more."

40-80% of hires come through referrals.

But I couldn't remember who I knew at which company. My network was data I wasn't using.

Source: Cultivated Culture, Glassdoor, Wall Street Journal

SECTION 02

The approach

If you had a magic wand, what would you want?

I asked myself a UX question

"If you had a magic wand, what would you want?"

Monitor target companies

Score jobs for fit

Match contacts to roles

Draft outreach messages

Tailor resumes automatically

Run while I sleep

Everything I just described became my first prompt to Claude.

The tool stack

n8n Cloud

Automation engine

Notion

Databases (Jobs, Companies, Contacts, Materials)

TheirStack API

Job data from 55 companies

Claude API

Message and content generation

OpenAI TTS

Audio briefings

Gmail API

Email delivery

Built with Claude as my architect and debugger. No engineering background required.

SECTION 03

The system

9 workflows. 88 hours. One goal.

System architecture

Each workflow runs automatically, passing data to the next.

5:40 AM

1. Auto fetch jobs

Pulls jobs from 55 companies via TheirStack

5:45 AM

2. Score jobs

Calculates fit: tier, salary, remote, contacts

5:45 AM

3. Match contacts

Links 440+ contacts to open roles

On Decision

4. Generate materials

Warm and cold outreach, or resume/cover letter

6:00 AM

5. Morning briefing

AI summary + audio delivered to inbox

On Error

6. Error notification

Alerts me if anything fails

55

companies

450+

contacts

4

databases

5

API integrations

Note: V1 architecture shown. System has since expanded to 9 workflows including Company Intelligence Briefs, Gumroad product packaging, and error classification.

SECTION 04

Human-AI co-design

How a non-technical designer built a technical system.

The build process

Claude as architect, debugger, reasoning partner, and colleague.

Plan — Define goal, data needs, failure cases

Build — One step at a time, with screenshots

Break — Test until something fails

Fix — Debug with Claude, repeat

KEY INSIGHT

Failure isn't rejection, it's data. Each failure taught me something about how the system worked.

What I kept human

Trust and security design means keeping humans in control of decisions that affect other humans.

AI HANDLES (~85%)

HUMAN CONTROLS (~15%)

Decision to reach out or apply

Review every message before sending

Review every resume before submitting

Final judgment on tone and fit

"The system generates drafts, not finished products. The final call is always mine."

SECTION 05

The outcomes

From chaos to control.

Before vs. After

Monthly savings: ~12.5 hours (at 10 applications) | System cost: ~$70/month

The hidden benefits

The unexpected impact on my process and mindset.

Dread

Anticipation

Avoidance

Engagement

Chaos

Control

Guilt spiral

Clarity

Decision fatigue

Data-driven decisions

"I actually look forward to checking my job search now."

SECTION 06

Public validation

Building in public. Validated in public.

LinkedIn series performance

8 posts documenting the build process over 3 weeks.

TOTAL REACH

29K+

impressions

288

reactions

AUDIENCE QUALITY

35%

Senior-level professionals

15%

Enterprise (10K+ employees)

ENGAGEMENT

9+ meaningful DM conversations with designers wanting to build similar systems

Tina Huang (AI/ML YouTuber): "Your AI agents video inspired my job search automation"

Podcast feature

Unsolicited invitation based on LinkedIn visibility. Episode link: https://youtu.be/utJ3wNGrTZQ?si=zwlTm8H_cQXHguv7

"This is SO cool Renee! I hope we see more job seekers taking your lead and leaning into the power of AI for the search. Amazing work :)"

— Austin Belcak, commenting on Post 1

Invited to walk through the full system on the Cultivated Culture podcast.

AUSTIN BELCAK'S AUDIENCE

1.4M+

LinkedIn followers

250K+

newsletter

9.52

satisfaction /10

Founder of Cultivated Culture

​

Helps people land jobs without applying online

SECTION 07

The workflows

For those who want to see under the hood.

Workflow 1: auto fetch jobs

Pulls job data from 55 companies daily at 5:40am via TheirStack API.

Workflow 2: scoring relevant jobs

Calculates fit score based on tier, salary, remote status, and my 440+ contacts

Workflow 3: matches contacts & links to companies

Identifies who I know at each company and links them to open roles

Where the magic happens

In Notion, the Jobs database stores new job roles, scores for fit and matches contacts in real time

Workflow 4: generate outreach & application materials

Three paths: Referral messages for warm and cold outreach, and/or resume/cover letter generation

Generated referral messages

In Notion, the Outreach database produces warm and cold outreach messages organized and ready for my review before sending

Generated application materials

In Notion, the Application Materials database creates tailored resumes and cover letters with built-in quality evaluation

Workflow 5: morning briefing

AI-generated summary + audio briefing delivered to my inbox at 6:00am

The morning briefing

AI-generated summary + audio briefing delivered to my inbox at 6:00am daily

https://bit.ly/renee-ai-briefing

Workflow 6: error notification

Alerts me immediately if any workflow fails

SECTION 08

Beyond v1

What I shipped next.

What I built next

Closing the gaps I identified in v1.

V1 system: Discovery → Scoring → Outreach ✓

Gaps I identified and closed:

This is what it looks like to apply multi-agent decomposition: identify the full user journey, ship an MVP, evaluate what's missing, and iterate.

RR

I'm an AI Systems Designer, not an engineer.

If I can build this, so can you."

Let's connect.