News for hackers. Stuff that matters.

jlewi

Re:Foyle: You build it, AI should run it on Thursday October 17, 2024 @10:23AM (#41869958)
> Is there a difference with a set of scripts that have been carefully reviewed and approved

I think Foyle is complementary. Creating a script/tool/higher level abstraction is valuable when you do the same exact task over and over. This is creating a so called "paved path". However, paved paths often limit flexibility. To optimize flexibility we often divide a task into composable tool chains. Foyle can help you compose these chains. So it can help you in instances where it doesn't make sense to create a bespoke tool because it wouldn't be used frequently enough.

(I'm the creator of Foyle)

Re:Foyle: You build it, AI should run it on Thursday October 17, 2024 @10:16AM (#41869904)
The premise of Foyle is that you can solve this problem with runbooks that walk you through the steps to deploy your app; in this case setting up appropriate networking. Problem is no one wants to read or write documentation. So Foyle tries to learn by observing what commands people do. So in this case, an expert (e.g. someone on the infra/platform team) deploys an app or troubleshoots a networking problem and they do this by creating a notebook which contains some exposition and the commands. Foyle learns from this. So now when an app developer goes to deploy a new app, they start a document "Deploy Caribou" and Foyle starts suggesting the steps and commands to deploy it.
Re:Foyle: You build it, AI should run it on Wednesday October 16, 2024 @07:26PM (#41864924)
Being able to review a command isn't the same as being able to author that command without errors. An obvious example is queries (e.g. SQL or Cloud Logging). I can easily know enough to be able to review them and decide a query is safe. That doesn't mean I can easily remember the precise syntax that each tool/system requires.
Re:Foyle: You build it, AI should run it on Wednesday October 16, 2024 @06:20PM (#41864421)
I'm the creator of Foyle. Why do you assume the user lacks the competence to review the commands? In order to execute the commands you need the relevant tools (gcloud, kubectl, etc...) to be installed and you have sufficient privileges. Why would you install and grant privileges to tools you don't know how to use safely?
Hey all,

I'm the engineer at Google leading our Kubeflow effort. Happy to talk to anyone interested in think about contributing or using Kubeflow.