Elissa

Blog Post: So You're Forced By Your Work To Use LLMs To Program...

https://goodnameforablog.com/posts/using-llms-to-program-unfortunately/

When You Have to Use LLMs to Program, Unfortunately

Over the last few months especially, I find myself frequently asked an awful a lot of questions about using LLMs to help program. My first instinct was to glibly say “don’t use them”, it which is fine especially for all the ethical reasons and environmental reasons… but unfortunately for many people who are stuck with jobs that they need in order to pay the rent, this is not an option.

Good Name For A Blog
Feb 12 at 12:44amWeb
Show threadKyle BashourFeb 12
@vampiress @joe I am not an LLM shill, in fact I have always been highly skeptical. Nor am I someone in management trying to get more people to adopt it for some reason. But I just wanted to suggest - if you're not "directly connecting it to your code base" you've not given an it an honest go. These are immutable statistical models *except for what you put into the context window* and tools like Claude Code or Codex are massively more capable than chat bots because of this.
Show threadKyle BashourFeb 12

@vampiress @joe And specifically with the latest frontier models, starting with Opus 4.5. Tool use and subagents for gathering context and iterating on changes, along with good prompting, make for a completely different kind of experience than chatting.

I'm a staff engineer at a fintech that is looking to become a bank. We have no interest in writing poor quality code. I've been amazed at how much more I've been able to do over the last few months with these tools.

Show threadKyle BashourFeb 12
@vampiress @joe I completely agree with point #2 - however, again, if you use tools like Claude Code, you *can* use feedback to directly improve the LLM. CLAUDE.md and agent skills have an absolutely massive impact on the quality of their code and relevance to your codebase, and give the tool memory.
Show threadSoulful Minion of UnorthodoxyFeb 12
@vampiress I really can't see how this would make it better or easier on my soul. I'd rather anger my boss than start using AI.
Show threadDr. RakunFeb 12

@vampiress So You're Forced By Your Work To Use LLMs To Program, malicious compliance edition:

1. Distance yourself from your shit job where you're not even trusted enough to pick your tools for yourself
2. Prompt a lot of agents, do xkcd sword battles while they're running
3. Superficially and very slowly edit the results for plausible deniability
4. Blame the LLMs for the resulting drop in productivity and quality
5. You will be fired, but it is their goal anyway. Enjoy your stressless life