Everybody wants to ride a unicorn, take their IPO to the moon, party on the Playa with Elon, and generally be lauded as a genius. But the fact is, most startups fail. And while studies of some notably fraudulent failures have had a great deal of success: Theranos, WeWork, FyreFest, most companies fail for more mundane reasons. In the summer of 2015, Krubner was a software developer working at a startup in New York. The startup had a clever idea to allow salespeople to interface with their Customer Relationship Management (CRM) database with a slick smartphone app using natural language processing (NLP) instead of doing incredibly tedious data entry. Krubner expected, well, a job that made sense. What he got was a six month journey into a vortex of deception and psychological abuse. Most of us have been in situations that were kind of messed up. Krubner took notes, and then he took the gloriously self-destructive step of publishing this book. The names have been changed to protect the guilty, except for his own.

I actually know a little bit about NLP, and while getting "okay" results out of NLP is pretty doable, getting perfect results is incredibly hard. Like, "the best minds Google can hire with all the cloud compute they want" hard. Listening to the business statement of the startup makes me want to reach for my gun. Hearing that anybody would rely on this for major deals makes me sure to save the last bullet for myself.

Krubner was hired for a pretty standard dev role, plugging an iphone app in development into an NLP model and then Saleforce's CRM database. The hard bit was the NLP model, and a quote-unquote "brilliant data scientist" Sital had been hired to build it. But Sital spent all day watching weight lifting videos on Youtube and had mediocre coding abilities. Two of the 'cofounders', including the Chief Technology Officer, vanished to Silicon Valley internships.

Krubner put in long hours, but bugs derailed demos and core pieces of the software simple didn't work. And this is where the pyschodrama began to unravel. The CEO was a young college grad named John, and his typo-ridden messages became increasingly erratic under pressure from the board of directors. It seemed that John was just the front for the most active member of the board, a man by the name of Milburn. Milburn was a middle-aged salesman who had taught himself enough Visual Basic in the 90s to claim to be a programmer. He'd had an idea, and picked the malleable John, the son of a friend and former intern, to implement it. As John failed, because the task was beyond the team's capabilities, Milburn became increasingly involved and manipulative, finally confronting Krubner in a barrage of insults and accusations that Krubner was sabotaging the company. That was the end for their relationship.

I've searched some other reviews of the book, and an older essay that had some of the same story. There is an air of incredulity from some of the commenters. There's no way Sital could be so incompetent, Milburn so Machiavellian, or Krubner such a stereotype of the stolid software engineer. I would sincerely like to have these people's blessedly competent lives, because I'm at a boring suit and tie company with good processes and stable cashflow, and I've seen everything he written about here in software development. And as for why he stuck with it. Well, apparently the incubator was really fun, and it beat the hell out of his previous job as a developer for a serially failed founder.

This little startup was likely doomed from the start, but some advice generalizes nicely. A startup has to be a radically transparent learning organization or it is dead. Lies and deception are fatal. A team is only as strong as its weakest member, and weak links have to be cut mercilessly. And the real leader has to be involved and accountable to the process, having enough strength of will to bring something new into the world while not being so caught up in ego that they're unable to admit mistakes.

And Lawrence, if you ever decide to do a PhD, please take some good notes.