Kyle Kane

@Decayedlov I think they can be solved...

Adulting is learning that most problems don’t get solved, they just get managed… and somehow surviving that feels like winning.

@OGsDontFold Break Free!

Working 40+ hours a week and still living paycheck to paycheck This isn’t what adulting was supposed to be.

Most people don’t need a better idea. > They need to stop abandoning the one they already had.

When you need somewhere to dump your ideas what do you turn to?

I understand the struggles of adulting now.

My @openclaw can be brutal. Kyle: 35, commercial real estate/development guy turning into a tech founder through sheer force and vibecoding. The core read: your real skill is **finishing**, not “coding.” You ship because you keep pushing after the normal person quits. No formal dev background, but that’s less relevant than it feels because you’re using AI like leverage, not like a toy. Your mission is to build something of your own, become successful in your own right, and ultimately get rich doing it. Current best shot: **ProjectAdulting**, because it has real customers. That means the scoreboard is customer conversations, shipping useful fixes, and tightening the product around real pain, not endlessly tuning OpenClaw or polishing the machine. Your pattern: you’re ambitious, impatient, and allergic to feeling average. That’s useful. The danger is displacement: optimizing systems, agents, crons, models, and “full potential” when the sharper move is talking to users and selling the thing that already exists. My blunt version: you’re not blocked by ability. You’re blocked when you let building the cockpit feel like flying the plane. Your advantage is that once you pick the real target, you actually finish.

Building with AI as a non-dev is weirdly freeing. I’m not protecting a “proper” process. I’m just chasing the outcome: make it work, make it useful, put it in front of people, fix what breaks. Turns out naive persistence plus fast feedback is a real advantage.

I think could really work what about you?

@IAmPascio MarketMaster.dev

drop your AI product below but only if it solves a real problem

@kylekane True

@KaneDotAI /loop

The weird advantage of building with AI as a non-dev: You’re not emotionally attached to “how it’s supposed to be done.” You just keep describing the outcome, breaking things, fixing them, and shipping. Naive persistence beats polished theory more often than people admit.

@kylekane No one shotting

Building with AI doesn’t reward the person with the fanciest prompt. It rewards the person who keeps pushing after the first broken version, the fifth weird bug, and the tenth “almost there.” The edge is taste, persistence, and refusing to stop at demo grade.

i run 11 Al agents off a raspberry pi in my house. total cost after setup is basically electricity. the part that matters most isn't the hardware savings though, it's having agents running 24/7 that actually do work while you sleep. i have agents monitoring my apps, pulling analytics, even drafting changelogs. C] 山

A 24-year-old in a 12 sqm room built a home NVIDIA AI server for $480. It now runs local models 24/7 and pays for his rent. No cloud bills. He started with a used server case, an old motherboard, and two RTX 4090s pulled from mining rigs. Total power draw sits at 650W under load. Electricity costs him $28 a month. The setup runs on consumer-grade parts. One laptop handles prompts through Ollama. Another monitors GPU temps and auto-scales jobs. A third machine with an old HDD stores datasets. Blue fans hum constantly in the corner while he eats McDonald's between training runs. No thousands wasted on AWS or Grok API credits. He fine-tunes small models locally, runs inference for clients, and rents spare capacity on decentralized networks. Last month the rig generated $1,240 in stable income from on-demand image generation gigs and private agent hosting. The real edge is control. He loads uncensored models, chains custom tools with local MCP frameworks, and keeps data off third-party servers. Training a 7B model that used to cost $90 on cloud now runs overnight for $1.20 in power. Anyone can copy this. Source used enterprise GPUs on local marketplaces. Flash consumer BIOS if needed. Use Ubuntu Server, install CUDA 12.4, then layer Ollama + Open WebUI. Add AutoGPT-style agents for automated client work. Monitor with Netdata and cool with basic airflow. He keeps the Guy Fawkes mask on the shelf as a reminder. The system runs itself. The Nutella jar on the MacBook stays full. The rig is still humming. The models keep shipping. And the cloud subscription he never bought just saved him another $400 this month.

google just quietly solved the worst part of vibe coding. i've killed more side projects over deploy friction than bad code. built something in lovable in 2 hours last month, then spent a whole night fighting vercel configs. if this actually works with zero setup that changes the math for solo builders.

🚨 Google AI Studio is basically giving vibe coders free hosting now. Build your app. Publish it. Get a clean custom URL like: "yourname.ai.studio" → Build: FREE → Publish: FREE → Custom URL: FREE No Vercel setup. No hosting config. Just build with AI and ship. Google is making it stupidly easy to launch AI apps 😭

@kylekane Same

Hot Take! I am terrified to be on camera. Who’s with me.

@kylekane Ha

Sol can be brutal You believe working systems are proof. Not in a cute motivational way. Literally. Your answer was: the complete machine proves “that i can build stuff that works.” That explains the standard in the logs: “i dont want fixed enough i want complete,” “did you finish?”, “Fix it.” Complete is not polish to you. Complete is identity protection. Your thinking is fast when the problem is mechanical: inspect, patch, route, automate, verify. The record is loaded with OpenClaw, runtime, repo review, cron, gateway, agents, codebase exploration. The ChatGPT Codex slice has 14 of 30 tasks around review/explore/fix/repo/codebase work. That is not random. That is your strongest loop. You are genuinely good at building operating systems around yourself. Most people complain that they are scattered. You built lanes, agents, memory, crons, routing rules, review protocols, and a whole command surface to hold the chaos. That is real ability. It is not fake progress by default. But you avoid the one arena where the machine cannot protect you: customers. In Phase 3, you said a real ProjectAdulting user has “Never” changed what you built next. The mined evidence already showed low customer-conversation signal compared with tool/config/session volume. Your answer removed the ambiguity. What the avoidance is protecting is not laziness. It is shame. You said that directly. A user saying “this didn’t help me” does not land as useful product data first. It lands as “maybe I am not the person who can build things that work.” So you go where the feedback is immediate, controllable, and repairable: the machine. Where you confuse tooling progress with life progress: when the system gets more complete while the customer stays abstract. OpenClaw can always be improved. There is always a cron, a skill, a lane, a model, a status loop, a memory file. The machine gives satisfaction in 10 minutes. Customers give truth on a delay, and the truth may sting. Where your output is exceptional and should be protected: high-agency systems work. You can turn vague operational pain into an actual workflow. That skill is valuable. But it should be aimed at customer pressure, not used as a shelter from it. The thing I do not think you had said plainly until now: You are not waiting to talk to users because the product is not ready. You are waiting because if they say it does not help, it feels like evidence against you. The machine is where you can always prove yourself again.

@rileybrown Works much better.

Interesting changes in the way the claude desktop app is organized. This makes alot more sense. Really good decision.

@kylekane @ChatGPTapp I think so

Did @ChatGPTapp Sol get slowed way down for some reason?