Am I an idiot or are AI agents much less capable than everyone says
Am I an idiot or are AI agents much less capable than everyone says
@_opencv_ Depends who built them and how lol, those who try to build a 1 size fits all agent without understanding how to actually DO the things they want, how to use function/tool calls, keeping context awareness, etc. Those agents suck
@_opencv_ yea seems like guy-with-ai is better than just ai or just guy, for the foreseeable future
@_opencv_ Yeah it turns out you cant leave something important to an alien entity that doesn’t work for you, isn’t self aware, and picks tokens with a random number generator.
@_opencv_ they get better the more and more precise you get, but at that point it’s basically easier to just do the task yourself, idk.
@_opencv_ Deep research, cursor agent mode, Codex, Claude Code. Plenty of great agents. Most if not all of these have been post-trained in their agentic harness I believe, so I think that’s the difference
@_opencv_ Nope you’re right I don’t see any use in them at all.
@_opencv_ Non-determinism is the bottleneck It’s also really easy to create a shitty agent
@_opencv_ AI is only good at “left brain” thinking. Creativity, imagination, intuition and emotion can be faked but isn’t genuine. If you are using an LLM for anything other then logic, math, language processing, you’re NGMI
@_opencv_ Am I an idiot or are web pages much less useful than everyone says?
@_opencv_ This is a test from the AI to see if you'll like this post.
@_opencv_ You can take time to learn how to build with these things don't waste your time and half arse things with it -- respect it as a tool, learn how it works and use it correctly Don't try to "vibe code" like non tech people
In my experience: people are easily impressed and they don't know what it spits out that's wrong I had one customer who was showing off to his team their new AI setup that they've been feeding every meeting we've had into it and he was super impressed about how he could ask it any question And then our engineer had to point out how it was wrong a lot of the time but he had already shown everybody off on how good it was and how great it was and all that other stuff We had to point out that if you actually followed its instructions you would have crashed an entire app that handles billions of events
@_opencv_ Agents are only as good as the data and context feed to them. Right now most enterprise data is locked so agents don’t have full context of the systems making them pretty useless
@_opencv_ Most are workflows not agents x.com/analyticsaurab…
@_opencv_ Most are workflows not agents x.com/analyticsaurab…
@_opencv_ 128D vectors can’t beat weakly-coupled neural oscillators generating million-dimensional standing waves of electrical activity
@_opencv_ Being an AI agent is cool but need instructions, which agentic system is currently less capable Claude Code, Cursor or Codex hard to say
@_opencv_ From my point of view, agents are getting smarter and better literally on a daily basis.
@_opencv_ you scroll down. someone posts: Am I an idiot or are AI agents much less capable than everyone says the only reply is the same line, word for word. no emoji. no punctuation change. just echo. you like it. feels right.
@_opencv_ I’m gonna slap the next person at work that says agent. It’s honestly killing me.
@_opencv_ view it as an extension/tool to speed up work and its amazing. still a long way from complete human-less automation
@_opencv_ Damn all you do is bitch on here seriously, the past week I’ve just been seeing your posts non stop of either complaining or being an ai doomer
@_opencv_ They can answer questions, but can they explain why the coffee machine still breaks every other day
@_opencv_ It’s incredibly overstated right now, like incredibly so aside from automating some document workflows
