How we Built an Automated Job Hunt Workflow with n8n, AI, and a Lot of Debugging
Like most creative folks in tech right now, We’ve been riding the wave of reinvention, exploring what’s next, prototyping wild ideas, and trying not to scream at Google Sheets. This is the story of how we built our own job application automation using n8n, OpenAI, and some friendly robot co-pilots.
The Idea: What If Applying to Jobs Didn't Suck?
Designing our own job hunt system came from burnout. Customizing resumes, tracking applications, and losing focus is a big problem in 2025. We wanted a workflow that works with AI to help users move fast, stay creative, and make job applying easy.
We used n8n to automate customizing resumes and cover letters using a job description link.
n8n has a small learning curve if you know node-based interfaces. Our first workflow took several hours to build but will save a lot of time later.
Each node in n8n does a task, connects services, and passes data to the next node until the workflow is done. In this case the final step was recording recording basic information about the role, and creating custom content for the resume and cover letter. But more importantly we learned how to build workflows in n8n which should set us up with a repeatable process and a starting point for future workflows.
This is an example of the n8n interface, using my resume update automator.
What we Built
HTTP Request: Grabs the job description from the link.
OpenAI Node 1: Extracts key role requirements and necessary skills.
OpenAI Node 2: Generates a tailored resume summary (voice- and tone-matched matching our own)
Wait Node: Keeps the workflow human (and avoids API throttling).
Google Sheets Append: Logs job title, link, summary, and timestamp.
🤖 Stack Snapshot
n8n: open-source automation builder
OpenAI GPT-4o-MINI: used for job description parsing and summary generation
Google Sheets: job tracking and storage
HTTP requests: to scrape job post content
The Debugging Spiral (aka Syntax Hell)
Things were going great — until they weren’t. 😤
The Google Sheets node refused to accept my data. I spent hours in expression purgatory:
Invalid expressions (red), not valid ones (green)
No data passing between nodes
Every syntax combo I tried failed.
Cue existential dread.
Enter the AI Assistants
What saved us? The n8n AI Assistant (Beta) 🧠
Claude kept repeating the error, but n8n’s built-in assistant spotted it right away. It had more context into the actual nodes (even if it couldn’t read screenshots), and immediately flagged the issue: OpenAI was returning a JSON string, not a structured object. I needed to manually parse it.
Rookie mistake, now fixed. A good reminder that AI works best when paired with human intervention — the kind that knows when to pause, question the output, and bring in the right kind of help.
The exact fix looked like this:
🤖 Claude vs n8n Assistant
Both tools were essential — but for very different reasons.
Claude | A strategic thinker that accepts screengrabs. Great for workflow design, prompt engineering, and debugging logic. |
n8n Assistant | Precise and contextual but doesn't read screengrabs. Understood the node structure and fixed syntax quickly. Can't read screenshots or external syntax. |
They weren’t redundant — they were complementary. And when both hit a wall, we did the most human thing possible: went back to a previous version and rebuilt from there. Ironically, that ended up being faster.
It was a solid reminder that contextual AI — the kind built into the tools you’re already using, is the future. We won’t always need to bounce between standalone models like Claude or ChatGPT. Eventually, your design tool, workflow builder, or spreadsheet will come with its own embedded copilot, one that knows your data, your process, and how to actually help.
🚀 Final Result
Fully automated, AI-personalized resume summaries
All logged and tracked in Sheets
Scale-ready for bulk job apps (and maybe, someday, recruiter alerts)
✨ Creating with AI isn’t about mindlessly automating everything. It’s about knowing when to let the machine run and when to step in as the human.
AI helped us format, parse, and fly through dense job descriptions, but it was our human instinct that spotted the generic prose, fixed the syntax bugs when Claude and n8n assistants kept repeating the same mistakes, and knew when to hit “restore previous version” instead of spiraling.
Sure, the custom resume summary and cover letter are pretty generic — but they’re a solid starting point. And instead of spending 90 minutes staring at a blank doc, I got there in 15 seconds. Think of it as a glorified outline. From here, it’s up to you to refine, shape, and inject your personality. That part’s still very human work — and thankfully, still out of AI’s hands (if it had any). ✋
Here are some pro-tips:
Use AI like a team: Claude for high-level thinking, n8n Assistant for precision.
Debugging is still a skill, but one that’s easier when shared with smart tools.
When the AI assistants are stuck, they are missing something outside the context, and sometimes that means an easier way to do it.
Sometimes being human is just knowing the simpler way.
Thinking about turning this into a shareable template or tool, would that be useful to folks? if so, reach out!
