What does it mean that optimize the bottleneck first?

Improving anything other than the bottleneck is wasted effort.

Example: You run a weekly content pipeline: research on Monday, drafting on Tuesday, editing on Wednesday, publishing on Thursday, promotion on Friday. Each step takes roughly a day except editing, which routinely spills into Thursday and sometimes Friday, delaying publication and eliminating promotion entirely. You decide to optimize the pipeline. You could speed up research by subscribing to a better source aggregator. You could speed up drafting by using templates. Both are genuine improvements to real steps. Neither will change your output. Your pipeline publishes one piece per week — sometimes zero — and that rate is determined entirely by the editing bottleneck. If you cut research time in half, you finish Monday's work by noon. Then you wait. The draft still arrives Tuesday. Editing still overflows Wednesday. Publication still slips. You optimized a non-constraint and the system did not notice. Now instead, you focus on editing. You create a style guide that eliminates recurring decision points. You batch similar edits. You establish a 'good enough' threshold that prevents perfectionist loops. Editing now fits within Wednesday. Publication happens Thursday. Promotion happens Friday. Output doubles — not because every step got faster, but because the one step that determined system throughput got faster. Every hour you spent optimizing research was wasted. Every hour you spent optimizing editing was leveraged.

Try this: Map a system you operate — a workflow, a daily routine, a project pipeline, a learning process. List every sequential step. For each step, estimate the time it takes and whether downstream steps must wait for it. Identify which step most frequently causes the rest of the system to wait. This is your current bottleneck. Now list the last three improvement efforts you made to this system. For each, answer honestly: did you improve the bottleneck, or did you improve a non-constraint? If you improved a non-constraint, estimate how much system-level throughput actually changed. The answer is likely close to zero. Finally, generate three specific actions that would reduce the duration or increase the capacity of your actual bottleneck. Rank them by effort-to-impact ratio. Execute the highest-ranked action this week and measure whether system throughput changes.