Only three kinds of AI products work

Only 3 AI products work? Internet yells “prove it” as devs dunk on chatbots

TLDR: A viral take says the only winners are chatbots, code autocomplete, and coding agents. Commenters clap back, adding image/music tools and non-chat AI, mock “chat-with-tools” as a jailbreak risk, and want real workplace helpers—while devs joke the “agent wrote the tests” so don’t trust the pass.

An essay declared there are only three AI hits: chatbots like ChatGPT, code “autocomplete” like GitHub Copilot or Cursor, and hands-on coding agents (think Claude in IDEs). The comments immediately turned into courtroom drama meets roast battle. One camp cheered the “keep it simple” verdict; another yelled objection: you forgot the booming image/video/music crowd and all the AI that doesn’t look like a chat window. “This list is biased toward chatty bots,” critics said, arguing real wins hide in less flashy tools that quietly do work. The spiciest subplot: trust. Devs mocked the idea that it’s “easy to verify” AI code because, as one put it, the agent sometimes wrote the tests—aka the fox grading its own homework. Meanwhile, the article’s dunk on “chatbots with tools” (the refund-me $999 jailbreak meme made the rounds) had the thread chanting: buttons beat bots. People want one-click actions, not “dear diary” dialogs to change a font size. On the saucy side, the “AI companions” niche got side-eye, with folks predicting big labs like xAI’s Grok and OpenAI eventually push into more adult-friendly territory. Office warriors chimed in begging for real workplace agents for SharePoint and Office, not another chat bubble wearing a tie.

Key Points

  • The article asserts most AI products are chatbots and only a few categories truly work.
  • AI labs have advantages—access to newest models and co-developed harnesses—making their chatbots superior to third-party offerings.
  • Explicit roleplay chatbots succeed commercially due to fewer content restrictions, though the niche faces ethical issues and potential competition from major labs.
  • “Chatbots with tools” (AI assistants) are deemed ineffective due to jailbreak risks and inferior chat UI compared to direct product controls.
  • Completion-based coding tools like GitHub Copilot and Cursor Tab are presented as effective, delivering immediate value by smart autocompletion.

Hottest takes

"This is actually a low bar, when the agent wrote those tests" — 8organicbits
"Author forgot about image, video, and music creation" — Shebanator
"This seems to be biased heavily towards products that look like an LLM" — wongarsu
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.