I build ML/AI systems for money since 2019 and NOTHING works like the hype posts say. NOTHING. Am I the only one sick of this?
retained text
Every week my feed has a new version of the same post. You know the one. \> It has never been easier in history to build a $10-100k/month business. I physically can't keep up with all the opportunities I see. \> My agents manage agents that manage agents. 30+ subagents running for two days on one task. I don't read what they do, reading that slop is pointless, I only talk to the manager. \> Half my agents are automations: they review every new deal, answer my email and Telegram, analyze traffic, set my goals for the week. Plus the reels. "I built a gorgeous 3D website for a local business, my team of AI agents did everything." "My whole business is AUTOMATED by agents." Said with a completely straight face. A lot of those are paid collabs with the AI tool in the video, and the rest have a course link in the description. And then you look at what the author actually makes money on. A course about AI. A paid channel about AI. Consulting on "AI-native transformation". Deal flow for a VC fund. So the endless opportunities are... selling AI to people who are scared of being left behind by AI. Man, that's a business model for sure, just not the one they're advertising. And if it's not a course, it's a FACTORY. Content factory. GTM engine. AI SDR that books meetings while you sleep. "100 faceless reels a day on autopilot". "Ship a website / an app from one prompt". AI ghostwriter that writes your LinkedIn posts AND comments under other people's posts. Ok, think about it for a sec. If everyone has a factory pumping out 100 posts a day, who's reading them? Other factories? Half of LinkedIn already is bots writing "Great insights!" under bot posts, and cold email is AI SDRs pitching AI SDRs. Every post in the feed is worth less, so you need MORE posts, so you buy a bigger factory. And the only one in this loop who's guaranteed to make money is the guy selling the factory. **Before you call me a luddite**. I build this stuff with my own hands, for money. Two examples. I ran an AI persona that posted fully autonomously on Instagram and Twitter **for a year and a half.** Actually autonomous, nobody writing the posts for it. And surprise, the metrics kept slowly growing the whole time. So yes, I know autonomy can work. I also know how much work went into designing how that character thinks before it posted anything worth reading. And it was ONE character with a thought-through head, posting like a person, not 100 reels a day. I built a multi-agent system that screened hackathon applications for a small VC fund. EVERYTHING went wrong. Every step. Yes, models were weaker back then and there were no proper harnesses, I know. In the end it worked: analyzing all applications from one hackathon went from 200+ hours to 4.5. Real business result. And it took a team of three AI engineers three months. Nine engineer-months to save \~200 hours per hackathon. That's what "AI automated it" looks like in real life. And in all this time NOT ONCE did something work on the first try. Every single time it's: run it, read what it actually did, find where it went off the rails, fix, run again. So when someone tells me "my agents review every new deal in the background, I don't read what they do", I have questions. Who checked it? How many good deals did it throw out? How many of those auto-replies in your email were wrong? If you have those numbers, PLEASE post them, I'm serious, it would be the most useful AI post of the year. If you don't, how do you know it works at all? And ofc you may say: dude, that was old models, now it's spec-driven development, the AI writes the product spec, then the tickets, then builds the whole thing. Ok. A good chunk of my paid work is taking apart products built like that. Sites and apps that are supposed to solve a real problem, made by someone who supposedly knows exactly what UX they want, and product-wise they are as bad as it gets. You've seen them. Pills with text inside. Three-layer headings: tiny all-caps eyebrow in IBM Plex Sans, giant headline, gray subheadline. Don't take my word for it. Describe your project in as much detail as you want and throw it into Claude Design, Gamma, Lovable. You'll get garbage. Decks too. If AI agents make your decks from one prompt, please, just show me one. Either that deck solves nothing and you didn't need it, or... yeah, there's no second option. A good slide is ONE thought plus the PROOF of thle rule. And not one tool does it. Not Gamma, not anything else Itried. TEXT TEXT TEXT TEXT. Text everywhere. I make decks with AI myself and it's NEVER one prompt. About "never been easier to build a business". Yeah, building got cheap. But distribution was always the hard part, and now it's even harder, because every weekend MVP and every factory output is fighting for the same eyebe who already had an audience, a network and money. So yeah,easiest time in history, for people who were already doing fine, and for everyone else it's the same grind with way more noise. So, honest questions: **1. Has anyone here built something with AI that makes real money and is NOT selling AI courses, AI consulting, AI factories or AI tools to other AI people?** **2. If you run multi-agent setups for real work, how do you check tit?** Or is it just me who's sick of this? TL;DR: AI works, I build with it. It never works from one prompt, hs, and the loudest "my agents run my business" people mostly sellcourses and factories. Show me the numbers.
document fetched 2026-10-10 00:28 · canonical https://reddit.com/r/OpenAI/comments/1x1nxom/i_build_mlai_systems_for_money_since_2019_and