I'm personally not convinced. I mainly use LLMs for research I would have otherwise done using search and vibe coding experiments (so that I can properly address people / pull requests in the job).
For the former, I do think that good prompt formulation helps, and personally I should give much more thought to my prompts, because I find I have to redo them too often. However, I do worry that if we're de-naturalizing language too much in this interface, that the goal of having a natural language interface is shifting: we're adopting to inferior tech?
For vibe coding stuff, now that I've built a framework to instruct the bot how to handle things, and depending on the implementation used of course, I find that I can short-hand prompt and do most of the work outside of the prompt in issue management. Specifically with Claude, prompting becomes "propose a fix/solution for issue/feature request <a>" or "implement the proposal from issue <a>", which technically I could trigger automatically. I can't help but feel that with persistent comms (like on github between humans) this is a longer term viable solution for software development than doing everything in a prompt.
Ironically one of the best prompting tips that seems to work for me is prompting as if you're talking to a junior team member, a little more touch and verbosity really helps with the fact that these things are at their core just autocomplete. Without the consequence of confusing someone we tend to treat them like a CLI, or at least I did.
@Grok is this true?lol<a>" or "implement the proposal from issue<a>", which technically I could trigger automatically. I can't help but feel that with persistent comms (like on github between humans) this is a longer term viable solution for software development than doing everything in a prompt.