The fix was two lines. Finding it took a day and a half. That ratio is the whole job.

A few weeks ago I found a bug that only showed up under load, three services deep, in code I had not touched in months. The fix was two lines. Finding it took a day and a half.

That ratio never changes, and it is the most honest description of senior engineering work I know. Writing code is the fast part. Understanding why it broke is the slow part, and the slow part is where engineers actually separate.

The chase looked like every real chase does. The error surfaced nowhere near the cause. The obvious theory was wrong, which cost half a day, because the obvious theory is always wrong in exactly the way that makes it look right. Reproduction required load, so every experiment was slow. The breakthrough was not clever. It was instrumentation placed where the evidence said to place it, not where intuition wanted to look, and then believing what the data showed even though it contradicted the mental model I had walked in with.

None of that day and a half was typing. It was building a mental model of a system across three services, generating theories, killing them with evidence, and updating honestly. That skill has a name, judgment under uncertainty, and it is built one painful chase at a time. There is no shortcut, no course, and no tool that does it for you.

Which is exactly why the current moment is widening the gap between engineers, not closing it. AI made the fast part faster. It has barely touched the slow part, because the slow part requires context no tool holds: what the system is supposed to do, what changed recently, which past incidents rhyme with this one. The engineers who were already good at the slow part now get the fast part nearly free, and compound. The engineers who only had the fast part just watched their differentiator become a commodity.

If you are deciding what to get better at, get better at the chase. It is the part that appreciates.