Comment on Amazon Allegedly Replaced 40% of AWS DevOps With AI Days Before Crash

just_another_person@lemmy.world ⁨4⁩ ⁨months⁩ ago

• FreedomAdvocate@lemmy.net.au ⁨4⁩ ⁨months⁩ ago

  Dev Ops is one of the most automated parts of software development and deployment actually.

  source

  • just_another_person@lemmy.world ⁨4⁩ ⁨months⁩ ago

    …noooooo, it most definitely isn’t.

    While the job does deal heavily in automating things, it only automated Boolean things. Looking at a platform and seeing why and where it’s failing is not a Boolean thing, and never will be. It’s the same reason we still don’t have machines that repair cars over 100 years after their introduction.

    source

    • AnarchistArtificer@slrpnk.net ⁨4⁩ ⁨months⁩ ago

      Exactly. DevOps engineers are already super skilled at using automation where appropriate, but knowing how and when to do that is still an extremely human task

      source
    • FreedomAdvocate@lemmy.net.au ⁨4⁩ ⁨months⁩ ago

      Looking at a platform and seeing why and where it’s failing is not a Boolean thing, and never will be.

      AI can see why and where it’s failing too if it has the appropriate permissions and access.

      It’s the same reason we still don’t have machines that repair cars over 100 years after their introduction.

      No it’s not. DevOps is all software, repairing cars is not. Car ECUs can tell you exactly what is wrong with your car.

      source

      • just_another_person@lemmy.world ⁨4⁩ ⁨months⁩ ago

        🤣

        Okay bud, go and tell AWS, Google, Salesforce, and any other of these companies who think “AI” is an answer to everything, because they’ve all had very public outages due to this exact same thing in the past few months.

        You have no idea what DevOps is or how it works if you think any of this is easily done or solved with these junk tools.

        source

        • -> View More Comments
  • cley_faye@lemmy.world ⁨4⁩ ⁨months⁩ ago

    Automation with a lot of validation steps that are not very obvious. Because if they were, we’d have automated them away.

    source
  • kcuf@lemmy.world ⁨4⁩ ⁨months⁩ ago

    Devops is often figuring out why automation didn’t work.

    source

    • FreedomAdvocate@lemmy.net.au ⁨4⁩ ⁨months⁩ ago

      Absolutely, but that’s not an argument against what i said.

      source

      • kcuf@lemmy.world ⁨4⁩ ⁨months⁩ ago

        I feel like it’s directly contradicting what you were implying…

        source

        • -> View More Comments
  • Amir@lemmy.ml ⁨4⁩ ⁨months⁩ ago

    Tried to figure out yesterday why a user couldn’t ssh into a server, tried LLMs to figure it out, completely useless. Had to go into some log file somewhere to find out the one who set up the server made a specific group for ssh and if a user wasn’t in that group they couldn’t connect. The LLMs (ChatGPT and Gemini) gave me bullshit about changing flags in the sshd config…

    source

    • cute_noker@feddit.dk ⁨4⁩ ⁨months⁩ ago

      Easy fix: give an LLM root access to all production critical servers and allow everyone in the company to chat with it.

      source

      • qarbone@lemmy.world ⁨4⁩ ⁨months⁩ ago

        First reaction: fear.

        Second: I chuckled. Because I thought of some VP-level enforcing this joke as SOP.

        And then a little more fear, as a treat.

        source

        • -> View More Comments
    • FreedomAdvocate@lemmy.net.au ⁨4⁩ ⁨months⁩ ago

      Cool story. LLMs can’t see error messages in logs if you don’t give them access to it. Had you given the AI agent access to those files? Or were you just using a standalone LLM with no access to the system?

      source
  • squaresinger@lemmy.world ⁨4⁩ ⁨months⁩ ago

    DevOps is not executing the automation, but designing it. DevOps is not manually spinning up pods but writing the automation that does so.

    source

    • FreedomAdvocate@lemmy.net.au ⁨4⁩ ⁨months⁩ ago

      Why do you think AI can’t easily be the supervisor for pipelines and create new ones? It’s basically just creating steps that are well known from building a branch to deploying it.

      source

      • squaresinger@lemmy.world ⁨4⁩ ⁨months⁩ ago

        Try it out and you’ll see. Amazon seems to be doing great with it.

        source

        • -> View More Comments
  • sabin@lemmy.world ⁨4⁩ ⁨months⁩ ago

    Most of my work in DevOps isn’t in front of my text editor writing scripts. It’s spent hopping between dashboards, drafting emails, doing RCA, teaching dev team members how to use pipelines, and getting requirements from them for designing new pipelines. Then inevitably debating with them about design considerations when they ask for a set of procedures that won’t pan out.

    Until your AI is a fully fledged team member who everyone can feel comfortable engaging with as if they were a real human, you cannot possibly begin to automate this.

    source

    • FreedomAdvocate@lemmy.net.au ⁨4⁩ ⁨months⁩ ago

      Most of my work in DevOps isn’t in front of my text editor writing scripts.

      Mine either - we use azure devops, octopus, etc - tools that automate devops. Most of devops is automated across the board - no big companies are manually kicking off builds for every PR and pushing the files around the place and then manually deploying them - it’s all automated using devops tools. Having AI build and manage these pipelines seems like a logical place to use it, as they are all just about creating steps using pieces from previous steps and other systems.

      You absolutely could have AI create a pipeline to build, test, and deploy a solution, and then test the actual deployed solution. The AI is essentially just the coordinator here, tying together the other devops tools.

      source