Here's the full stack (and the business hiding inside it) Last week Grok Bot cut everyone's usage limits. I run 8 bots on it. A chief of staff, a Skool moderator, a Slack drafter, a YouTube comment bot, a Bookface outreach bot, and a few more. Overnight, all of them were rationed. So instead of waiting for xAI to fix it, I tried something different: I had my agents reverse engineer Grok Bot and build me my own. It worked. I recorded the whole thing, every prompt, start to finish. Video is below. Here's the breakdown. What actually happened 1. The design. I pointed Claude Opus 5.5 at the real Grok Bot app on my Mac and told it to click through every screen with computer use, then rebuild it in Paper MCP. About 30 minutes later I had a near pixel-perfect replica: same screens, same icons, same layout. 2. The backend. In parallel, a second agent ran the Reverse Engineer Anything repo (443K stars on GitHub) on a Linux cloud computer and mapped how Grok Bot works: the bots, the API, the interface. 3. The build. The build agent read the design thread, read the architecture, and built it. A couple of hours from my first prompt, I had a working clone. Google sign-in, Grok 4.7 behind every bot, app connectors, routines, and each bot on its own cloud computer that I can take over and control. While all that ran, a third agent on a GPU box edited the video you're about to watch. My MacBook never got warm. The stack - T3 Code: where the agents live. One workspace running Claude and Codex across 5 machines: my MacBook plus Linux, GPU, Mac mini and Windows cloud computers. - Reverse Engineer Anything: maps how an existing app works. - Paper MCP: design tool for agents. This is what nailed the UI. - xAI API: the models, voice, image and video, files, X search. Most of what Grok Bot is. - Composio: sign-in and every app connector (Gmail, Slack, all of it). - Orgo: a computer for every bot. (Yes, that's my company. It's also the piece xAI doesn't give you.)

Watching a computer-use agent click through an interface on video makes automated operations look almost solved. The friction usually starts the moment you try handing that exact workflow over to run reliably inside a messy client setup. Most of the real challenge seems less about setting up the agent logic and more about keeping track of edge cases without sitting there watching it execute. For those running these for real businesses, where does the workflow usually break down first?

I just built a full AI employee from scratch in this video. Everyone’s hyped on agents but barely anyone’s actually showing how to build ones that can really work inside a business. Here’s the entire stack I spun up live (no holding anything back): • Cloud computer running Hermes on Orgo • Its own email (AgentMail) • Its own phone number (AgentPhone + iMessage) • Telegram as the main chat interface • Every tool & connector hooked up through Composio • Credit card so it can actually spend money • Obsidian vault as its knowledge base / second brain • Latitude for observability so I know when shit breaks I made one main orchestrator agent (named it Hubert) that stays in charge 99% of the time and just hands tasks off to specialized sub-agents underneath it. One giant bloated agent is a nightmare to debug. This way everything stays clean and purpose-built. Full live build, every prompt I pasted, every terminal I opened… it’s all in there. And I’m giving away the complete templates for the orchestrator + every sub-agent I build going forward. I'm giving you actual employees that can do real work. Watch the whole thing here: PS - the template is free here on Github: https://github.com/nickvasilescu/nicks-stack

When I first started exploring AI agents, I thought the hardest part was finding the right tools and getting everything connected. I spent so much time looking at different platforms, testing ideas and trying to understand how everything worked. At some point, I realized I was focusing more on the technology than the actual problem I wanted to solve. What changed things for me was learning to start small. Instead of trying to build a complete AI employee, I started looking at one simple business task that could be made easier through automation. That taught me something important: you don't need the most complicated system to create value. You need a solution that solves a real problem and works consistently. I'm still learning, but that shift in thinking has made the journey much clearer. For those just getting started, what's your biggest struggle right now: choosing the right tools, building your first agent, or finding your first client.

just dropped the episode with greg isenberg. if you only watch one video from me, make it this one. the idea: take a personal agent platform like grok bot, niche it down to one industry, and install it into a business as a full team of agents. each agent gets its own computer, email and phone number. the owner can text it, call it or message it in slack. i built my version (bops) in one weekend. 36 hours, one long thread in t3 code with opus 5.5. it's open source, so you can fork it and skip the weekend. everything below is covered in under an hour. the build process: 1. start with research so the model understands what you're building 2. design it all in paper before writing code. ask for 3-4 variations, then merge the best parts 3. point your coding agent at the docs and API reference for every tool 4. screenshot bugs and UI you like, then feed them back in the stack 1. orgo for the computers 2. agentmail for email 3. agentphone for phone numbers 4. composio for app connectors 5. host on orgo, fly or vercel the offer 1. $25k upfront: half at signing, half after a 30 day implementation 2. optional $3k/mo to maintain it. to the owner, that's like hiring a team of employees at $12 an hour 3. they own the system, so they never feel stuck on a retainer 4. add a free website redesign guarantee results you know you can hit, like faster replies to leads, or refund the month you can also go one-to-many: build it for a single industry and sell it as a product. niche pricing can run 10-20x what grok bot charges. getting leads 1. don't pick a niche yet 2. go through your phone and call 50 people who run a business or know someone who does 3. do the first one free for a case study, and ask for 3 intros to their peers in return sit with that first client, watch how they work, and build agents for what you see you never sell cold this way. every call comes through someone who already knows you. 🎥 video: https://www.youtube.com/watch?v=67BmkptgTms