Ask HN: Are AI Agents a Lie?

2•David1238•8mo ago

LLMs awesome. They’ve changed how I code (Cursor), search for things, take notes, brainstorm, write emails, etc. (ChatGPT).

But I don’t use any AI agents. My friends don’t either. I haven’t even heard of any company building AI agents outside of the annoying customer service ones.

By AI agents I mean an LLM that does things in the background without needing constant feedback. Like booking a flight or scheduling a meeting.

The only useful “agentic” quality I’ve seen is basic file or web search to better answer my questions, which doesn’t feel like an agent.

Am I missing something? Why is there so much hype? Am I just understanding agents wrong?

Comments

AsmodiusVI•8mo ago

They are not a lie but useful agents that can take actions are still relatively rare. At MS Build this week I saw one that can write notion summaries of GitHub issues. I also saw one that actually resolved GitHub issues assigned to it. There’s lots more out there and little by little you’ll see them show up. The addition of MCP will help here as it gives “agents” the ability to connect to things and take action but within limits, therefore less likely to break stuff.

TheMongoose•8mo ago

https://www.wheresyoured.at/ has a lot to say about the topic.

jasonthorsness•8mo ago

IMHO most companies touting AI agent capabilities today are just participating in the hype cycle to get attention and funding. But don't let that lead you to think AI agents will never actually work. They are inevitably going to happen as a simple evolution of non-AI bots that already exist and have proven applications.

How capable and flexible agents will become and how cost-effective they will be is completely unknown. Cost is going to matter a ton. I'd love to just have an agent that constantly watches dashboards for me powered by o3 but I'd be bankrupt before the end of the day.

collingreen•8mo ago

It's about trying to find a business model with high value so you can do "value based pricing". If you can replace a human salary you can charge a lot. If you are seen as just a chat bot, you can't.

AI Watermark and Stego Scanner

Clarity vs. complexity: the invisible work of subtraction

Solid-State Freezer Needs No Refrigerants

Ask HN: Will LLMs/AI Decrease Human Intelligence and Make Expertise a Commodity?

From Zero to Hero: A Brief Introduction to Spring Boot

NSA detected phone call between foreign intelligence and person close to Trump

How to Fake a Robotics Result

It's time for the world to boycott the US

Show HN: Semantic Search for terminal commands in the Browser (No Back end)

The AI CEO Experiment

Speed up responses with fast mode

MS-DOS game copy protection and cracks

Updates on GNU/Hurd progress [video]

Epstein took a photo of his 2015 dinner with Zuckerberg and Musk

MyFlames: Visualize MySQL query execution plans as interactive FlameGraphs

Show HN: LLM of Babel

A modern iperf3 alternative with a live TUI, multi-client server, QUIC support

Famfamfam Silk icons – also with CSS spritesheet

Apple is the only Big Tech company whose capex declined last quarter

Reverse-Engineering Raiders of the Lost Ark for the Atari 2600

Show HN: Deterministic NDJSON audit logs – v1.2 update (structural gaps)

The Greater Copenhagen Region could be your friend's next career move

Do Not Confirm – Fiction by OpenClaw

The Analytical Profile of Peas

Hallucinations in GPT5 – Can models say "I don't know" (June 2025)

What AI is good for, according to developers

OpenAI might pivot to the "most addictive digital friend" or face extinction

Show HN: Know how your SaaS is doing in 30 seconds

ClawdBot Ordered Me Lunch

What the News media thinks about your Indian stock investments