📄 Article
✓ Link checked
Free
Intermediate
Why we picked it Walbaum ran ops at Uber, Meta, and Rippling, so he writes from the exact judgment call in the question: when has the manual work done its job. His answer is a loop (hit a new problem, build a scrappy manual fix, extract the learnings, then automate) with the signal being when the manual solution becomes a real bottleneck, not before. It is the practical counterweight to Graham's essay.
How to win by doing things that don't scale
From The Operator's Handbook by Torsten Walbaum
- Automate when the manual approach becomes a genuine bottleneck, not on a schedule and not because it feels unglamorous.
- The manual phase exists to extract learnings, if you productise before you understand the problem you build the wrong thing and waste the weeks.
- A concrete UberEats example (SQL and Python scripts before a real product) shows what "validate manually first" looks like in practice.