alt Hacker NewsI’m seeing a lot of people that having that train of thought. I think that is a fallacy. I can understand that some LLM execution can be faster, but definitely not all. Agents need to explore, grep and get back up to speed to get context, if you have a good mental model, you can do changes or adaptions in <7s with a bunch of shortcuts or commands.
One could argue to find the <7s commands in your head takes you more mental power than to just wish it in to the LLM and whilst its running you can wish something else in another session, but I’m thinking that the cost of context is more important than the actual execution time for your task. Every extra task gets more expensive. It’s not a ressource where you have a limit, right from the 2nd task the cognitive load increases.
Therefore I’m thinking one task that can be done in one context window without switching is worth a load in these days of constant distraction.
I get your point but I do prefer delegating almost all of the things to the LLM.
A lot of times human commands are prone to errors / edge cases as well. Example a simple git pull command usually would take < 7 seconds, but then LLM can take care of resolving merge conflicts etc as well.
A simple git push is usually instant but that comes with an overload of some un intended changes being pushed, which LLMs take care of removing themselves.
When LLM is doing its thing, we can spend the same time in writing the next prompt.