Ask HN: Which AI concepts are here to stay, and which will churn?

4 points by datsci_est_2015 6 days ago | 7 comments

There’s a multitude of concepts that have been created during this advent of AI. Which do you predict will be resilient and long-lasting, and which do you think will churn away as harnesses evolve?

  - Model Context Protocol
  - Skills
  - Agentic workflows
  - Agents
  - Claude.md / Agents.md
  - etc.

Bonus: Despite all of the above, what do you see as the unaddressed or “dark corners” of the AI ecosystem?

JadAmmar 5 days ago |

In my opinion, these B2B AI assistants aren't going to save companies that much money (at least not as much as people think). I also think the future of popular AI software lies in apps that don't require users to interact with an LLM at all. To me, it feels like AI is best placed in the background, invisible, but working to assist us in our daily tasks.

kordlessagain 6 days ago |

All will churn for a while. This is a fundamental shift in how we use computers and all the things connected to them. Automate everything is going to occur sooner than later.

One "dark corner" is agent to agent payments. When that takes hold, agents will be paying other agents to do what they are good at, and the ones that accel will make bank and spread.

We are unlikely to survive as-is. We too must evolve.

moomoo11 5 days ago | |

it will be just like the current trend too.

your agent pays the other agent, it doesn’t really work. maybe 10k agents do this.

the other agent will book that revenue maybe 100k in a month, cool they’re 1.2M arr. raise 20M, revenue drops drastically as number of suckers drops.

eventually the company will wind down and the next grift will start.

kordlessagain 5 days ago | | |

I do think there is value in some of it. If it moves to crypto, liquidity will settle where the value is. At least that is my hope.

lemonademan 6 days ago |

I believe Agents and the Model Context Protocol (MCP) will remain resilient, while rigid Agentic workflows will likely churn away. Hard-coded, step-by-step state machines—like rigid "if-this-then-that" prompt chains—are just temporary scaffolding. In my opinion, autonomous agents are likely going to be the most resilient piece here, but I also believe they harbor the most unaddressed "dark corners" of the AI ecosystem. A lot of people are rapidly deploying them to multitask like scheduling meetings, running and testing code, etc but there is also a massive lingering issue regarding how much power and execution authority we are giving them, and the unchecked decisions they could make could be problematic depending on the situation.

kordlessagain 5 days ago | |

It seems a lot of what they "release" is prompting, which is good for token burn.

meagher 6 days ago |

Not on your list, but getting annoyed with streaming text. Just want to see the message when it is complete instead of getting blasted by large number of tokens rendering per second.