Max Levchin

We have a whole culture of postmortems at a firm where we record every decision we make, especially I'll give you like a very retail one, because now that I said postmortem, postmortem culture comes from engineering. And obviously I'm an engineer by training and past. And so it's easy to reason in those terms.

We have a whole culture of postmortems at a firm where we record every decision we make, especially I'll give you like a very retail one, because now that I said postmortem, postmortem culture comes from engineering. And obviously I'm an engineer by training and past. And so it's easy to reason in those terms.

We have a whole culture of postmortems at a firm where we record every decision we make, especially I'll give you like a very retail one, because now that I said postmortem, postmortem culture comes from engineering. And obviously I'm an engineer by training and past. And so it's easy to reason in those terms.

But years ago, we had decided to very rationally scale our infrastructure because we expected a giant launch with just a crazy amount of transactions. And it was all going to be 10x bigger than what we were at the time. And so we were prepping at breakneck pace and scaled up every part of our backend infra and then our frontend infra and everything, everything.

But years ago, we had decided to very rationally scale our infrastructure because we expected a giant launch with just a crazy amount of transactions. And it was all going to be 10x bigger than what we were at the time. And so we were prepping at breakneck pace and scaled up every part of our backend infra and then our frontend infra and everything, everything.

But years ago, we had decided to very rationally scale our infrastructure because we expected a giant launch with just a crazy amount of transactions. And it was all going to be 10x bigger than what we were at the time. And so we were prepping at breakneck pace and scaled up every part of our backend infra and then our frontend infra and everything, everything.

And then the day off, it just fell apart massively because we basically overloaded our backend with our frontend. And so we were all dressed up with no place to go. And the entire database data layer of a firm was just in shambles for several hours while we were cleaning it up. But it really was a self-inflicted one where we were like, well, we just don't know how big it's going to be.

And then the day off, it just fell apart massively because we basically overloaded our backend with our frontend. And so we were all dressed up with no place to go. And the entire database data layer of a firm was just in shambles for several hours while we were cleaning it up. But it really was a self-inflicted one where we were like, well, we just don't know how big it's going to be.

And then the day off, it just fell apart massively because we basically overloaded our backend with our frontend. And so we were all dressed up with no place to go. And the entire database data layer of a firm was just in shambles for several hours while we were cleaning it up. But it really was a self-inflicted one where we were like, well, we just don't know how big it's going to be.

So let's assume it's going to be super huge. It was very modestly sized at launch and turned out to be just a giant sum. What do you learn from that? do the calculating. This was not an unforeseeable error. Someone should have sat down and said, all right, so we're scaling the frontier faster than we're scaling the back tier.

So let's assume it's going to be super huge. It was very modestly sized at launch and turned out to be just a giant sum. What do you learn from that? do the calculating. This was not an unforeseeable error. Someone should have sat down and said, all right, so we're scaling the frontier faster than we're scaling the back tier.

So let's assume it's going to be super huge. It was very modestly sized at launch and turned out to be just a giant sum. What do you learn from that? do the calculating. This was not an unforeseeable error. Someone should have sat down and said, all right, so we're scaling the frontier faster than we're scaling the back tier.

At some point, given the architecture of the system, you will have what's called an oversubscription of front end connections to the back end. And the back end will start choking because even if front end doesn't do anything, just holding these connections is expensive. If you have like a 10x mismatch, you're going to bring down the database. And that is what happened.

At some point, given the architecture of the system, you will have what's called an oversubscription of front end connections to the back end. And the back end will start choking because even if front end doesn't do anything, just holding these connections is expensive. If you have like a 10x mismatch, you're going to bring down the database. And that is what happened.

At some point, given the architecture of the system, you will have what's called an oversubscription of front end connections to the back end. And the back end will start choking because even if front end doesn't do anything, just holding these connections is expensive. If you have like a 10x mismatch, you're going to bring down the database. And that is what happened.

So someone should have done that work. that would have been a calculated risk that would not have backfired, but this time it did.

So someone should have done that work. that would have been a calculated risk that would not have backfired, but this time it did.

So someone should have done that work. that would have been a calculated risk that would not have backfired, but this time it did.

It's boring to analyze success. Like, well, we did well. It's good. I'm tired. I'm going to go to sleep is kind of the natural, like your oxytocin is in your brain.

It's boring to analyze success. Like, well, we did well. It's good. I'm tired. I'm going to go to sleep is kind of the natural, like your oxytocin is in your brain.