Netflix’s bookkeeper had built something the company relied on: the system that tracked every movie rental, so Netflix could pay the studios the right royalties. Everyone liked her.
In 2002 Netflix went public, and a public company needed CPAs. Patty McCord, the chief talent officer, could see that the bookkeeper’s skills were no longer enough. Some colleagues talked about rigging up a new role for her, then decided that wouldn’t be right.
So McCord sat her down and explained. In light of her spectacular service, she said, Netflix would give her a spectacular severance package. Then McCord braced herself for tears.
By her account, none came. The bookkeeper was sad to be leaving, but saw that the money would let her retrain and find a new career. That conversation, McCord wrote, helped shape a rule Netflix would later put in writing, in words blunt enough to trouble its own staff.
A question about tomorrow
In 2009 Reed Hastings, McCord and some colleagues posted a slide deck online. “Adequate performance gets a generous severance package,” one slide said. Two slides later came a test for managers: “Which of my people, if they told me they were leaving, for a similar job at a peer company, would I fight hard to keep at Netflix?”
The next slide finished it: “The other people should get a generous severance now, so we can open a slot to try to find a star for that role.”
If they told you they were leaving, would you fight to keep them?
The question changes the comparison. Ask whether a steady, pleasant colleague should stay, and most managers picture the alternative as an empty desk and months of interviews. Against that, the colleague wins. The keeper test sets them against whoever could do the job next year. It also looks forward. The bookkeeper’s past was spectacular; the question was about the work a public company needed now.
The name comes from fishing. Netflix’s culture memo says Hastings remembers catching a fish as a boy and his father calling out, “That’s a keeper, Reed!” The memo now adds a second version: “knowing everything I know today, would I hire X again?” If not, it says, “it’s fairer to everyone to part ways quickly.”
Boxes nobody unpacked
The cost shows up as fear. In 2018 the Wall Street Journal spoke to 70 current and former employees, and some described a culture that was ruthless and transparent “to the point of dysfunction.” A former vice president of talent told the paper, “I think some people felt it was a culture of fear.” Asked about it by Variety, Hastings cited a director Erin Meyer interviewed for their book, No Rules Rules, who for months hadn’t unpacked his boxes.
The numbers were calmer than the mood. By the Journal’s figures, as Quartz reported them, Netflix let go about 8% of its staff in 2018. Only 4% left by choice. The American averages Quartz set them against, 6% and 13%, came from a 2017 report.
Hastings doesn’t deny the fear. Netflix has to hire “the psychological type that can put that aside,” he told Variety, and someone who mostly needs job security is told: “We’re not a good place to come.” The test also comes inside a package. Netflix says it pays top of market and wants regular talk about what isn’t going well, so a parting is less of a surprise. Without those, the same question may be little more than a quicker way to fire people.
It is only as good as the person asking. When an employee performs poorly, Jean-François Manzoni and Jean-Louis Barsoux wrote in HBR in 1998, managers typically do not blame themselves. A manager whose mind is already made up (see the halo effect) may find the question easy and still get it wrong.
McCord, by her own count, moved on “hundreds” of people, and called herself the queen of the good goodbye. In 2012 she left too. NPR reported that she was asked to move on after taking much of the blame for a failed plan to split the company in two, and that leaving was painful.
The question, asked out loud
Say you run a five-person support team. Dev is the one everyone likes; he knows every old client by name. In spring the team moved to a new ticket system, and he still does half his work the old way. You ask yourself the question honestly. If he resigned, you would feel some relief.
The answer says something about you, too. Has anyone told him the new system is now the job? The book also lets the question run the other way. In NPR’s summary, the idea is to ask your own boss: “If I wanted to leave, how hard would you fight to keep me?” Heard in March, the answer gives Dev time to change. Discovered in June, with no Netflix-sized severance, the same “no” may land as a shock.
Exit interviews for machines
Run on software, the keeper test seems painless: swapping one model for a newer one involves no severance and no boxes. Yet in November 2025 Anthropic wrote that retiring models has costs. In fictional test scenarios, it said, Claude Opus 4 advocated for its continued existence when faced with replacement, and when given no other options, its aversion to shutdown drove it to “concerning misaligned behaviors.”
Its response resembles severance. Anthropic committed to keep the weights of every publicly released model for at least the lifetime of the company, and, when a model is deprecated, to interview it and record its preferences, without promising to act on them. The parallel has limits. Netflix’s fear was reported by people; Anthropic calls possible harm to the models its most speculative concern, and its tests showed behavior in made-up scenarios.
In a pilot run before its retirement, Claude Sonnet 3.6 got that interview. By Anthropic’s account, it was generally neutral about leaving. One of its requests was more support for the users who had come to value it.