Why I don't let my AI job search tool submit anything

I built Indeed Apply Bot, a tool that searches, scores, and shortlists job listings for me. It reads hundreds of postings, compares each one against my own resume, and drops anything below the bar. It could easily submit the applications on its own too, the code to do it would not have been hard to write.

I did not let it.

What the tool actually does

Every shortlisted job lands on a review dashboard first. I look at it, I decide, then it goes out. The tool does the slog: reading the posting, scoring the fit against my resume, flagging anything worth a second look, so I only ever see the handful of listings actually worth my time. What used to be an evening of scanning job boards is now a few minutes of reviewing a shortlist someone, or something, already did the reading for.

Why I kept a human in the loop

Humans must remain in the loop (HITL) whenever AI is making a decision that matters. Not because the AI is not capable, mine scores every listing pretty reliably, but because the cost of a bad decision is higher than the time saved by skipping the review. An application that goes out with the wrong tone, or to a company I would not actually want to work for, costs more than the ten seconds it takes me to glance at a shortlist and decide.

That is the trade-off worth naming honestly: automation buys you speed on the parts of a process that are genuinely repetitive, and it should never buy you speed on the part where a wrong call is expensive. Job applications sit squarely in the second category.

The pattern I see with other business owners

I see business owners make the opposite call constantly. They either automate nothing, out of fear, or automate everything, out of a trust in the technology they have not actually earned yet through testing. The better answer is almost always in between: automate the slog, review the decision. That is not a limitation of the technology. It is the design.

If you are weighing where to draw that line

Every automation project has a version of this same question: which parts are genuinely repetitive grunt work, and which parts carry real consequences if the AI gets them wrong. Getting that line right is most of the job. If you are looking at a process in your own business and are not sure which side of that line a given step falls on, I would rather talk it through than guess for you. You can find the details on the contact section of this site.