Back to blog
AI•May 14, 2025•4 min read

An apply pipeline that submits, but a human signs off

We built a pipeline that could tailor and submit job applications at scale. The tempting version of that project is full automation - let it rip and submit everything. The version that actually works draws a clear line between the toil, which the machine should own, and the moment of judgment, which a person should own.

A job application is a high-stakes, hard-to-undo action. You really do not want a bug quietly sending the wrong thing under someone's name a thousand times.

Automate the toil, gate the judgment

The toil here is real and worth removing: gathering the role, tailoring the application to it, formatting, queuing. The pipeline did all of that fast and consistently. But the actual submission, the irreversible part, sat behind a human sign-off. The person was not doing the boring work anymore, they were just making the call on something the system had already prepared for them.

Automate everything up to the point of no return. Put the human exactly at the point of no return, and nowhere less useful.

Make the review worth the human's time

A sign-off step is only valuable if the human can actually judge well in a few seconds. So the review had to show the right things - what is being sent, where, and anything that looks off - so the person could approve confidently or catch a problem. If you make the human review noise, they start rubber-stamping, and then you have the risk of automation with none of the safety.

How I think about human-in-the-loop pipelines:

  • Separate the toil from the judgment, then automate only the toil.
  • Put the human at the irreversible step, not at some arbitrary checkpoint.
  • Give the reviewer exactly what they need to decide fast and well.
  • Remember a rubber-stamp is not a safety net. Design the review so it stays real.

The pipeline saved an enormous amount of repetitive work, but it never pretended to have judgment it did not have. The person stayed in charge of the part that mattered.