3 ms·

amazon has mediocre quality talent that they grind to the bone. which worked when the company just needed raw execution. amazon has an operations culture, which

amazon has mediocre quality talent that they grind to the bone. which worked when the company just needed raw execution. amazon has an operations culture, which was important for: 1. scaling retail 2. keeping the servers running at AWS all the low hanging fruit has been picked, they need a fundamentally different employee base

spwa4 9mo ago
> all the low hanging fruit has been picked, they need a fundamentally different employee base As anyone in software development can tell you, this does not compute. You cannot do things this way, and any experienced software engineer can tell you it doesn't work. Besides, it's not how Amazon worked at all. Amazon is famous for having systematically verified ("mathematically proved") how it's core systems operated. Whereas, for example Google only did that in redesigns when the systems had already collapsed once or twice due to scale, not from early on. And even that is superior to how Microsoft or Oracle did it: they bought Google employees and had them design an iteration of what Google is running (yes, is running, not was running. Google redesigned it's core systems ... and then mostly didn't migrate. Borg was never replaced with Omega and the main large system that they migrated to is Spanner. Kubernetes isn't Borg. Kubernetes grew out of Borg's successor. Except Google never migrated away from Borg) https://cacm.acm.org/practice/systems-correctness-practices-at-amazon-web-services/ https://cacm.acm.org/practice/systems-correctness-practices-... I'm sure Amazon had entire departments, much larger than core engineering, just like every other company, where it looked like everything was operationally focused. That doesn't mean core engineering doesn't exist, or does nothing.
- vineyardmike 9mo ago
  Publishing a paper (in 2025) about formal verification is miles different than implementing it company wide. Anyone who has talked to an Amazonian SWE knows they’re not building their systems like this. The Amazon way is to quickly grind something out, and build on as many layers of AWS abstractions as possible. They’re famous for the hustle and grind not the stellar engineering acumen of formal verification.
  - spwa4 9mo ago
    Goes back quite a bit further: https://www.amazon.science/tag/formal-verification https://www.amazon.science/tag/formal-verification
  - geodel 9mo ago
    Agree. I'd say they may have build 1% systems like this. But impression is whole cloud is built with formal verification

spwa4 9mo ago

> all the low hanging fruit has been picked, they need a fundamentally different employee base As anyone in software development can tell you, this does not compute. You cannot do things this way, and any experienced software engineer can tell you it doesn't work. Besides, it's not how Amazon worked at all. Amazon is famous for having systematically verified ("mathematically proved") how it's core systems operated. Whereas, for example Google only did that in redesigns when the systems had already collapsed once or twice due to scale, not from early on. And even that is superior to how Microsoft or Oracle did it: they bought Google employees and had them design an iteration of what Google is running (yes, is running, not was running. Google redesigned it's core systems ... and then mostly didn't migrate. Borg was never replaced with Omega and the main large system that they migrated to is Spanner. Kubernetes isn't Borg. Kubernetes grew out of Borg's successor. Except Google never migrated away from Borg) https://cacm.acm.org/practice/systems-correctness-practices-at-amazon-web-services/ https://cacm.acm.org/practice/systems-correctness-practices-... I'm sure Amazon had entire departments, much larger than core engineering, just like every other company, where it looked like everything was operationally focused. That doesn't mean core engineering doesn't exist, or does nothing.

vineyardmike 9mo ago
Publishing a paper (in 2025) about formal verification is miles different than implementing it company wide. Anyone who has talked to an Amazonian SWE knows they’re not building their systems like this. The Amazon way is to quickly grind something out, and build on as many layers of AWS abstractions as possible. They’re famous for the hustle and grind not the stellar engineering acumen of formal verification.
- spwa4 9mo ago
  Goes back quite a bit further: https://www.amazon.science/tag/formal-verification https://www.amazon.science/tag/formal-verification
- geodel 9mo ago
  Agree. I'd say they may have build 1% systems like this. But impression is whole cloud is built with formal verification