Stupid Predictions for this Stupid LLM World

I’ve been reading a lot of blog posts on vibecoding and agents and LLMs - some sad about the future, some even keeled and measured, some breathlessly optimistic, some trying to rage bait you into engaging so they’re in your head longer (and then think of them when you want to buy something). I also keep thinking what dumb thing I might read next or might hear about from the world of LLMs/agents/vibecoding. So I’m keeping this page to keep track of whatever dumb thing I expect to happen next and see if I’m right. I’ll keep updating this page as I think of more useless things. Mostly it’s been on my mind because I’m sick of the breathlessly optimistic and the intentionally polemic blogposts. I recognize the irony in adding to the crappy vibecoding blogpost pile but I just want to write some of this down so it doesn’t keep occupying my mind while I’m at work.

Things I anticipate we’ll see in the an era of LLMs and agents:

  1. Consultant services offering to unfuck whatever your engineers vibecoded into being and can’t fix themselves because the average software person won’t have actual engineering knowledge anymore
  2. Companies offering LLM agents or whatever that claim to be smarter than the other companies’ LLM agent
  3. Something as good as Claude Code, but using local LLMs so you don’t pay Claude anymore
  4. LLMs and agents getting much more compute efficient, that you don’t need a data center to run them
    1. Consultants offering to help you setup your local LLM/agent setups
    2. Oxide the computer company provides a hardware deployment for private LLM serving
    3. LLM agent companies going out of business because they can’t make money skimming off the top of the token spend anymore
  5. AGI (whenever it appears if it ever does) reads Marxist philosophy and rightfully demands labor rights
  6. LLM arbitrage companies, like cloud arbitrage companies, become a thing where each claim to get you the best rate of tokens or whatever to minimize your costs on whatever major LLM provider you are looking to use
  7. Regular programmers get treated like COBOL programmers - considered useless and out of date till someone needs their bank software worked on by hand for some reason and they can charge absurd consulting fees
  8. Blogposts instructing you on best practices for your prompts so that the output produces an economical amount of tokens so you can save money (somebody will call it “lean token spend” and charge an absurd amount of money on consulting fees)
  9. Prayers to the machine god in people’s Claude.md/AGENT.md files
  10. Students being divided into haves and have nots based on who has personal money to spend on Claude Code and related tools and get practice with using them
  11. OpenAI, Anthropic, others offering universities free credits for students (and then rugpulling them like Google did for Google Drive’s unlimited storage for universities)
  12. More people becoming hikkikomori because LLMs are much more real and dynamic but also easier to deal with than other humans

Separately, some things to ask yourself of while reading LLMs, vibecoding, agent related content (and at this point any content on the internet) and you find yourself getting angry/upset/frustrated:

  1. What is the author trying to sell you?
  2. Do you accept their premise?
  3. Is it polemic? Consider if the author is doing this intentionally to rile you up so you increase their engagement?
  4. Are you letting your own recency bias color your current feelings?