> Claude deploy to dev will use abstractions to do it with software , even more it will deploy you a trigger to do it automatically
Who’s building the system that does that? Do you believe every request that we make - in every context that we make it - Claude will be able to intuit whether it’s something we only want to do once and is then just throw-away, or do it multiple times so a system should be built to facilitate that?
Even if Claude is somehow clairvoyant enough to know that, it still costs money to interpret my request in the first place. Why would I spend 5 cents to do something I could do for free? Maybe once, but across the day I’m doing hundreds or thousands of things and the money adds up. I’d rather have that money in my salary rather than having my employer pay for an AI subscription.
What’s the advantage of doing these things through a chat interface, aside from lining VC pockets? Vs clicking a button or invoking a CLI command? Just for the aesthetic of relinquishing control?
> Nothing wrong of doing it with voice / text request vs navigating to some ugly Jenkins , or half dead github actions etc .
I’ve automated away most of my daily browser activity using LLMs to generate tools. It’s become a lot easier recently. I do almost everything from a terminal, which makes it incredibly easy to e.g. pipe content into files for my LLMs to work with, like errors and stack traces. I understand some people have given access to their entire desktop to LLMs but that seems very solution-specific and I don’t want to commit days or even weeks getting my system to work with one piece of AI software just for another one to win the battle in the next few months. Too much churn. Text and terminal are evergreen.
Who’s building the system that does that? Do you believe every request that we make - in every context that we make it - Claude will be able to intuit whether it’s something we only want to do once and is then just throw-away, or do it multiple times so a system should be built to facilitate that?
Even if Claude is somehow clairvoyant enough to know that, it still costs money to interpret my request in the first place. Why would I spend 5 cents to do something I could do for free? Maybe once, but across the day I’m doing hundreds or thousands of things and the money adds up. I’d rather have that money in my salary rather than having my employer pay for an AI subscription.
What’s the advantage of doing these things through a chat interface, aside from lining VC pockets? Vs clicking a button or invoking a CLI command? Just for the aesthetic of relinquishing control?
> Nothing wrong of doing it with voice / text request vs navigating to some ugly Jenkins , or half dead github actions etc .
I’ve automated away most of my daily browser activity using LLMs to generate tools. It’s become a lot easier recently. I do almost everything from a terminal, which makes it incredibly easy to e.g. pipe content into files for my LLMs to work with, like errors and stack traces. I understand some people have given access to their entire desktop to LLMs but that seems very solution-specific and I don’t want to commit days or even weeks getting my system to work with one piece of AI software just for another one to win the battle in the next few months. Too much churn. Text and terminal are evergreen.