Claude Code + Paperclip: Running Companies With AI Agents

Julian Goldie just dropped a video about running an entire company with zero employees, and honestly? It made my brain do that thing where you have to pause and stare at the ceiling for a minute. 🤯

He's talking about combining two tools—Claude Code and Paperclip—to create what's basically a digital org chart filled entirely with AI agents. CEO agent, CTO agent, engineering agents, QA agents, the whole corporate hierarchy. Except nobody needs health insurance and the only thing that crashes is occasionally the code.

Let me break down what Goldie's actually demonstrating here, because this goes beyond "AI can write code now" (we been knew). This is about orchestration—making AI agents work together like a real team instead of just vibing independently in different browser tabs.

What Claude Code Actually Does

First, Claude Code. Most people still think of Claude as that really polite chatbot that writes your emails. Claude Code is different—it's Anthropic's coding agent that lives in your terminal and actually executes commands.

"It's not just a chatbot that writes code, runs inside a terminal, edits real files. It executes real commands," Goldie explains. "You give it a task, it figures out how to do it, and it does it like a real software engineer sitting at the computer."

Think of it like this: regular Claude is the friend who gives you a recipe. Claude Code is the friend who shows up at your house and cooks the meal. Both helpful, very different vibes.

It handles the full software development lifecycle—writing code, running tests, creating documentation, even deployment. It's not generating suggestions you then copy-paste. It's in there doing the thing.

Enter Paperclip: The AI Manager

Paperclip is where things get spicy. It's an open-source orchestration layer that lets you create an entire AI company structure. You literally build an org chart, assign roles, set budgets, and let the agents work.

Goldie describes it like this: "CEO makes decisions. CTO sets technical direction. Engineers write and ship code. QA catches bugs. Customer support handles requests."

It's giving very much "playing The Sims but everyone is an AI and they're building your actual business." 🎮

The key difference from just having a bunch of AI tools? Paperclip manages them like a remote-first company manages humans. Agents have budgets, can escalate decisions when they're stuck, and pass work between each other. It replicates that actual workflow where a junior engineer hits a problem, escalates to the CTO, gets direction, and implements the solution.

When you plug Claude Code into Paperclip as your engineering agent, you get what Goldie calls "a software engineer on your team who never sleeps, never complains, never asks for a raise."

How This Actually Works (The Part That Matters)

Okay, theory is cool and all, but what does this look like in practice?

Goldie walks through a real scenario: building an automated onboarding system for new community members. In the traditional world, you'd hire a developer, write a brief, wait weeks for them to build it, do revision rounds, maybe get something working eventually.

With Claude Code + Paperclip, you describe what you want in plain English. Paperclip spins up an AI company around that task. The CEO agent breaks it into a project plan. The CTO picks the tech stack. Engineer agents (powered by Claude Code) start building. The QA agent tests everything.

Here's where I need to pump the brakes on the hype, though. Goldie presents this as getting "a working product faster than you'd even finish the briefing document in the old world." Maybe! But that timeline depends heavily on how complex your ask is, how clearly you can articulate requirements, and how much debugging you're prepared to do.

The video doesn't show much actual debugging or problem-solving when things go sideways. And they will go sideways—that's just how software works. The real test isn't whether this system can build a simple onboarding flow. It's whether it can handle the inevitable weird edge cases and requirement changes that come up in any real project.

The Non-Technical Founder Angle

Goldie spends time talking about how this works for non-technical business owners, which is actually the most interesting part to me.

"You don't need to understand how the code works. You need to understand what you want the system to do," he says. "The skill with Claude Code and Paperclip is being able to describe the outcome you want clearly. That's it. The agents figure out how to get there."

This is genuinely significant if it works as advertised. Traditionally, if you wanted custom software, you needed to translate your business problem into technical language for developers. That translation layer is where a lot of things get lost or distorted (anyone who's ever worked with developers knows this pain).

Being able to describe your business problem in plain language and have AI handle the technical translation? That's a different game. But—and this is important—you still need to describe it clearly. Garbage in, garbage out applies to AI agents just as much as to human engineers.

The Reality Check Section

Let's talk about where this actually is right now, because Goldie does mention some caveats.

Paperclip is open source and moving fast, which means it's "not a polished one-click consumer product yet." You'll need some technical chops to get it running, or you'll need to find someone who does.

Claude Code is more accessible—it's a proper Anthropic product that's already available. But the real power comes from combining them, which means you're dealing with the friction of integrating two separate systems.

The budget management feature is actually really smart, though. Paperclip tracks what each agent is spending on AI compute and can make cost-aware decisions about whether tasks are worth completing. Because yeah, unconstrained AI spending will absolutely eat your lunch. Having the orchestration layer manage that automatically is genuinely useful.

What's missing from Goldie's presentation is the learning curve. How many iterations does it take to get your agent company running smoothly? What happens when agents misunderstand requirements or make architectural decisions you don't agree with? The real question isn't whether you can run a company with AI agents—it's how much supervision they actually need versus how autonomous they truly are.

What This Actually Means

Here's what I keep thinking about: we're watching the very early days of something that could fundamentally change how small teams operate. Not "replace all humans" (calm down, doomers), but genuinely shift what's possible for a small group of people to build.

The traditional constraint has been: you can only do what you have people to do. You need a feature built? Hire an engineer. Need customer support? Hire support staff. Your team size directly limited your capabilities.

If AI agent systems like this mature—and that's still a big if—that constraint starts to dissolve. A three-person team could potentially ship products that would've required fifteen people before. Not because the humans are doing more work, but because they're orchestrating AI agents that handle the execution.

That's exciting and slightly unsettling at the same time. The barrier to building software drops dramatically, which means more innovation but also more noise. The skill becomes less "can you write code" and more "can you architect systems and manage complexity."

Goldie's pitching this hard (he's clearly trying to get people into his paid community, which, respect the hustle), but the underlying tech is real and worth paying attention to. Claude Code is legitimately capable. Paperclip's orchestration approach is clever. Whether combining them creates the seamless AI company experience he's selling? That's something we'll only know as more people actually try to use this in production.

The future Goldie's describing might not arrive on the timeline he's suggesting—but it's definitely en route. And honestly? I'm here for a world where the bottleneck isn't finding good engineers, but knowing clearly enough what you want to build that you can explain it to an AI company.

The tools are getting wild, friends. The question is what we're gonna build with them. 🚀

—Yuki Okonkwo, AI & Machine Learning Correspondent