If you've offloaded execution to agents and quietly assumed the human variance problem went away, Dan Luu's essay is a useful cold shower to take this week.
Luu's "Individuals Matter" dismantles the comfortable managerial fiction that talent distributes so evenly across a team that no single person is truly load-bearing. He marshals case after case — from compiler work to hardware bring-up to early internet infrastructure — where one person's presence or absence bent the outcome dramatically. The lesson isn't that heroes are good; it's that pretending they don't exist is expensive self-deception that gets baked into org design, hiring, and retention until the company has quietly optimized for replaceability it doesn't actually need.
The agent layer changes what that irreplaceable person does, not whether they exist. In an AI-native studio the rare individual is less likely to be the one writing the most code and more likely to be the one who knows which agent outputs to distrust on Tuesday, which edge cases the eval suite systematically misses, or how to set task boundaries so the system doesn't confidently collapse toward a wrong answer. Luu's argument suggests you should be engineering your org to surface and retain that person, not structuring workflows that assume any reasonably senior human is a fungible node in a loop.
- Replaceability doctrine is a design choice that costs you more than you've priced in
- the irreplaceable person in an agentic org is often the one holding calibrated skepticism, not raw output
- find out who would genuinely hurt if they left tomorrow, then ask whether your retention and recognition structures reflect that reality.
