Why I stopped optimizing everything

For a while, I optimized everything: my morning routine, my inbox, my keyboard shortcuts, the title tags of pages nobody visited. If I did something twice, it got a spreadsheet. It felt like being very good at my job. Looking back, it was elaborate procrastination with charts.

Faster up the wrong hill

Optimization only works on the hill you’re already on. Every tweak gets you a little higher, and none of them asks whether you picked the right hill. SEO is full of this: re-optimizing an old article for the fifth time, or chasing a ranking for a keyword that stopped mattering, while the article your readers keep asking for never gets written. I’ve done both. The dashboard rewards you with a line that goes up, which is why it’s so hard to stop.

The questions that mattered most in my work were about direction. Should this page exist at all? Is this channel worth one more hour? Optimization can’t answer either one. All it can do is make the current answer more efficient, even when that answer is wrong.

Why a full calendar is slow

Queueing theory has a result I’d print above every calendar app. In the simplest model, a job takes five times as long to get through a system that’s 90% busy as one that’s 50% busy. At 99% busy, it takes fifty times as long. A machine shop booked to full capacity is a traffic jam that sends invoices. Every small surprise cascades, because it has nowhere to go.

My calendar was that machine shop. In a week planned to the minute, one sick day or one urgent client email knocks everything after it off schedule. So does a good idea that deserves an afternoon. Ideas took the biggest hit. Thinking, it turns out, happens mostly in the slack, the unplanned gaps between tasks. Take out the slack, and your work can only ever contain what you already knew.

A system with no slack looks efficient right up to the day it breaks.

What I still optimize

I didn’t swing to the other extreme. I just got pickier about where the effort goes. Three things still get the full engineering treatment:

• The few numbers tied directly to revenue. The rest is dashboard theater.
• Anything that repeats. A task you run fifty times a year earns its optimization; a one-off never does. The extreme case is the AI article pipeline I built at our agency. It has produced 400+ long-form articles, so every stage in it was worth tuning.
• Recovery time: how fast things get back to normal after a bad week. I trust it more than any productivity metric.

Everything else is allowed to be a little manual and a little slow. That’s a design decision, not negligence. I’d rather have a machine I still want to run in ten years than a perfect one.