If you automate the average, your edge is everything else.
Two signals point to the same thing: LLMs optimize for the average of their training, and the market for "average human tasks" is evaporating. For an engineer the conclusion is uncomfortable and clear: if your work is average, it's automatable. Your edge is everything that is not average — judgment, the exception, the hard call.
The thesis
An LLM gives you the mean answer of what already exists. At 3 a.m., in an incident, the mean answer is useless — what works is the judgment that sees the exception: the rare root cause, the non-obvious trade-off, the call under pressure with incomplete information. That can't be averaged. Senior value isn't knowing the common; it's recognizing the rare and deciding.
Signals of the week
Optimizing for the average kills the new
"Regression to the Mean: on LLMs and the quiet death of the new" — an essay gaining traction on HN argues that optimizing for the training average homogenizes and kills the unexpected. In engineering, the unexpected is exactly where judgment lives. Source: rruxandra.github.io.
The "average human" market evaporates
Amazon MTurk stops accepting new customers. The platform that defined micro-task crowdsourcing for 20 years is closing its funnel: demand for "cheap humans for simple tasks" dried up. Every average task is now an automation candidate. Source: TechCrunch.
My read
If your work fits within the average of the training data, it's a matter of time. But judgment —seeing the exception, deciding with incomplete information at 3 a.m.— can't be averaged. Don't compete where the machine is strong; double down where you're the one who decides.
Useful? Forward it to someone it'd serve.
Subscribe— Jorel