Installed Paperclip, Now What !?

My Paperclip article hit 2.7 million views. The comment section was mostly made up of "cool demo", "it doesn't work", and "how do you handle" questions... This follow up article is to get some ideas going in your head by using some really powerful skills. EVERYTHING is skills now... If you still aren't sure what Paperclip is and why the concept is so interesting than give the original article a review here 👇 Today we're going to stack skills... You'll need: Paperclip (the company), gstack (the engineering team), and autoresearch (the R&D lab). All free. All open source. Set up in under 10 minutes total. Real quick ... If you're non-technical and want to learn how to build systems like this, join our Build With AI community:http://return-my-time.kit.com/1bd2720397 Alright lets get going... Reminder, you're only limited by your own imagination.... This is just one (of many) examples... The Three Tools (And What Each One Does) Before we set anything up, here's the big picture. Paperclip is your company. It's the dashboard where you define your mission, hire AI agents, set budgets, and track what everyone's doing. Think of it as the office building. gstack is your engineering team. Built by Garry Tan (CEO of Y Combinator), it gives your agents 15 specialist roles - CEO, CTO, designer, QA engineer, release manager. Each one knows exactly what to do. Think of it as the employees. autoresearch is your R&D lab. Built by Andrej Karpathy (former Tesla AI lead), it lets your agents run experiments autonomously overnight. Give it a research question, go to sleep, wake up to 100 experiments completed. Think of it as the lab. Together: Paperclip runs the company. gstack builds the product. autoresearch does the research. One person. Zero employees. Step 1: Install Paperclip (2 Minutes) Open your terminal and run: This installs everything, creates a database, and opens your dashboard at http://localhost:3100. Once the dashboard loads: Create your company and set the mission (example: "Build an AI-powered note-taking app to $1M ARR") .. (make something better than that) Hire your first agent - it starts with a CEO The CEO will suggest hiring more team members Approve the hires and set monthly budgets for each agent Your company is live That's it. You now have an AI company with an org chart, budgets, and a ticket system. Step 2: Install gstack - Your Engineering Team (2 Minutes) This gives your agents the ability to actually build things. Run: git clone https://github.com/garrytan/gstack.git ~/.claude/skills/gstack cd ~/.claude/skills/gstack ./setup Now your agents have 15 specialist skills. Here are the ones that matter most: Planning your product: → /office-hours - Takes your rough idea and turns it into a real plan by asking smart questions and proposing three approaches → /plan-ceo-review - Challenges your scope like a CEO would. Finds the 10-star product hiding inside your feature list → /plan-eng-review - Locks down the architecture with diagrams, edge cases, and failure modes Building it: → /review - Code review that catches bugs your tests miss. Auto-fixes the obvious ones. → /qa - Opens a real browser, clicks through your app, finds bugs, fixes them, and writes tests so they don't come back Shipping it: → /ship - Syncs your code, runs tests, checks coverage, pushes to GitHub, and opens a pull request. One command. → /document-release - Updates all your docs (README, architecture, contributing guide) to match what you actually shipped Staying safe: → /careful - Warns you before anything destructive (deleting files, dropping databases, force-pushing code) → /freeze - Locks all files except the one folder you're working in. Prevents accidental changes during debugging. The power move: run 10-15 of these simultaneously. One agent does /office-hours on a new feature while another does /qa on staging while another does /ship on a finished PR. All at the same time. Step 3: Install autoresearch - Your R&D Lab (2 Minutes) This one is for when your AI company needs to experiment and learn. Run: Fire up claude code and ask it to make a skill called "autoresearch" based on https://github.com/karpathy/autoresearch.git. The skill builder will make you a version of Karpathy's ML training experiments. Autoresearch was built for ML training experiments, but the pattern works for any iterative research your AI company needs - testing prompts, optimizing workflows, benchmarking approaches. Here's how it works: → You give it a research question or experiment goal → It modifies code, runs a 5-minute training experiment, checks the results → If the results improved, it keeps the change. If not, it throws it away. → Then it does it again. And again. Automatically. → 12 experiments per hour. ~100 experiments overnight. You go to sleep. You wake up to completed experiments with clear results showing what worked and what didn't. How the Three Tools Work Together Here's where it gets powerful. Each tool handle