I think this idea applies beyond VIM and to text-based highly configurable tooling in general.
I'm a vim/tmux guy, and have been enamored with the new workflows possible with LLMs.
From the get-go, you can do things like:
`llm "Why is this build error happening? $(cargo build 2>&1)"`
It can write itself scripts and configs to trivially accomplish integrations:
`llm "Give me a new wrapper script for launching vim that uses the name of the current tmux window for its socket. Add an executable python script to ~/lib/llm/context/vim that uses the socket with the name of the current tmux window to dump all open buffers with headers indicating file name. Add another script to ~/lib/llm/context/tmux that dumps tmux pane scrollbacks with headers indicating pane numbers."`
Now you have a really useful coding assistant with minimal work:
`llm --context=vim,tmux 'tell me why I'm getting this compilation error'`
I had it build itself a tool to apply diffs to my files, so I can do:
`llm --continue --tool-cfg=editing 'Go ahead and fix that compilation error'`
I can have it build itself new tools.
`llm --tool-cfg=editing --context=toolbuilding 'Add a new toolset in ~/lib/llm_tools for running queries on the OpenSearch database running on localhost. One tool should dump the schema. Another should allow searching. Call the toolset config opensearch.yml. Write the backing script in python. Assume no auth necessary.'`
Now I'm able to use it:
`llm --tool-cfg=opensearch 'Are there any new records added to the incidents listings since yesterday?'`
llms aren't very good at solving novel problems in large-scale code-bases, so I'm not worried about them coming for my job, but they have made the annoying part of setting up my environment much easier! You still need to know how everything works so you can debug when it misses, but the quality of my dev tooling is now top-shelf and well customized to my projects with minimal investment.
I think this idea applies beyond VIM and to text-based highly configurable tooling in general.
I'm a vim/tmux guy, and have been enamored with the new workflows possible with LLMs.
From the get-go, you can do things like:
`llm "Why is this build error happening? $(cargo build 2>&1)"`
It can write itself scripts and configs to trivially accomplish integrations:
`llm "Give me a new wrapper script for launching vim that uses the name of the current tmux window for its socket. Add an executable python script to ~/lib/llm/context/vim that uses the socket with the name of the current tmux window to dump all open buffers with headers indicating file name. Add another script to ~/lib/llm/context/tmux that dumps tmux pane scrollbacks with headers indicating pane numbers."`
Now you have a really useful coding assistant with minimal work:
`llm --context=vim,tmux 'tell me why I'm getting this compilation error'`
I had it build itself a tool to apply diffs to my files, so I can do:
`llm --continue --tool-cfg=editing 'Go ahead and fix that compilation error'`
I can have it build itself new tools.
`llm --tool-cfg=editing --context=toolbuilding 'Add a new toolset in ~/lib/llm_tools for running queries on the OpenSearch database running on localhost. One tool should dump the schema. Another should allow searching. Call the toolset config opensearch.yml. Write the backing script in python. Assume no auth necessary.'`
Now I'm able to use it:
`llm --tool-cfg=opensearch 'Are there any new records added to the incidents listings since yesterday?'`
llms aren't very good at solving novel problems in large-scale code-bases, so I'm not worried about them coming for my job, but they have made the annoying part of setting up my environment much easier! You still need to know how everything works so you can debug when it misses, but the quality of my dev tooling is now top-shelf and well customized to my projects with minimal investment.
Considering vibe coding is the current state of the art I don’t think you have anything to worry about
Absolutely agree, trying to create a more user friendly site to learn vim called Vimgolf.ai
Eventually I plan on levels being ai generated and a bot being able to complete them to help users.