Back to blog
Engineering•June 22, 2026•4 min read

Manual is a smell: a four-step test before you automate

The fastest way to waste a month is to automate the wrong thing well. Before I build any pipeline, I run the work through four questions in order: what's the problem, how is it done manually today, what's the real solution, and which part actually becomes automation. Skip a step and you either automate toil that didn't matter or hand a machine a decision it shouldn't own.

Walk the manual path first

You can't automate a process you haven't watched a human do. The manual version is where the truth lives: which steps are repetitive, which are error-prone, which get quietly skipped when the week gets busy. That last category is the gold-mine-the high-stakes task nobody wants to do by hand is usually the one most worth making automatic, because automation doesn't get tired or forget.

Automate the step that's repetitive and high-stakes. Leave the step that needs judgment to a human-and make that boundary explicit.

Separate the toil from the judgment

The mistake is automating the part that looks impressive instead of the part that hurts. Grading essays looks like the hard problem; the weekly compliance audit is the actual pain. Once you've split toil from judgment, the solution writes itself: the machine takes the volume, the human keeps the call, and you've raised quality instead of just throughput.

A quick checklist before you build:

  • Problem: name the outcome, not the task. What goes wrong if this isn't done?
  • Manual: watch it done by hand once. Note what's repetitive vs. what needs a person.
  • Solution: decide what "good" looks like-faster, fewer errors, nothing skipped.
  • Automation: automate only the toil. Put a human gate on anything with real stakes.