Senior Product · Cloud Economics · Applied AI · $20M+/yr book · $1.3M+/yr saved · 60s→7sResume PDFPlain view
Back to Blog

Building a job-hunting agent that isn't allowed to lie

3 September 2026 · 1 min read

⇄ Paired with No. 004: Job Application Tailoring Pipeline

Most "AI resume tailoring" tools solve the wrong problem: they make a resume that sounds better, not one that's true. JobHuntAgent is the system I actually use for my own job search, and the constraint I designed it around is that it is not allowed to invent anything.

WHAT'S DONE

It runs as a Claude Code agent against a locked resume template: same fonts, same layout, only the content changes per job description. A one-time onboarding interview builds the reusable base profile once, and every tailored resume after that pulls from real bullets, reordered and reframed for what a specific JD cares about, never fabricated. A verification pass checks the output before it's allowed to ship, and there's optional tracking through Telegram, Gmail, and a small web dashboard so nothing falls through the cracks between applying and hearing back.

THE ACTUAL HARD PART

Writing a resume-tailoring prompt is easy. Writing one that refuses to lie under pressure, when a JD explicitly names a skill you don't have, is the actual engineering problem. That constraint is what makes the system worth building instead of just asking an LLM to "improve" a resume.

WHAT'S NEXT

More on the tracking side: turning "applied" into a real signal about what's converting to interviews, not just a list of jobs a resume got sent to.

Resume ↓Email