Paperclip: the open-source framework turning AI agents into companies
What's the deal? PaperclipDealroom has a profile for this one. Try Dealroom → is a new open-source framework that organises multiple AI agents into a functioning company structure. Built as a Node.js server with a React dashboard, it lets founders assign roles, budgets, and reporting lines to AI agents — turning tools like Claude Code, OpenClawDealroom has a profile for this one. Try Dealroom →, Codex, and Cursor into a coordinated workforce. One terminal command sets the whole thing up.
The project, created by developer @dotta and licensed under MIT, has already racked up over 26,000 GitHub stars and 3,000+ forks. It supports org charts, heartbeat scheduling, a ticket system with full audit trails, and per-agent monthly budgets that auto-pause spending at 100% utilisation. A marketplace called Clipmart is in the works, where users will be able to download pre-built company templates — a content agency, a trading desk, a dev shop — and run them out of the box.
Why now? The AI agent space is heating up fast. Microsoft just launched Copilot Cowork, built with Anthropic's technology. Anthropic shipped Claude Marketplace. Nvidia is building NemoClaw, an open-source agent platform. OpenAI is reportedly preparing its own cloud-based work tool.
But the real catalyst is a coordination problem. As agents got more capable, managing them didn't keep up. Running 20 Claude Code terminals at once means 20 disconnected sessions with no shared context, no cost tracking, and no way to recover state after a reboot. Paperclip's pitch: the bottleneck isn't the agents — it's the lack of organisational structure around them.
Several founders have already proven the "zero-human company" model works. Nat EliasonDealroom has a profile for this one. Try Dealroom →'s AI agent Felix has reportedly generated over $100,000 in revenue. Solo founder Aaron Sneed runs 15 custom GPT agents as a management council, saving him 20+ hours a week.
What could go wrong? Mistakes compound fast when agents feed outputs to each other. Australian AI consultancy Flowtivity learned this firsthand when a batch outreach hit 23 leads instead of three. Without strict governance — budget caps, human checkpoints, audit logs — costs can spiral and errors propagate before anyone notices.
There's also a security dimension. OpenClaw, one of the primary agents Paperclip orchestrates, has been flagged by Cisco as a significant security risk due to broad system permissions and susceptibility to prompt injection. Layering more autonomy on top of that raises the stakes.
The signal: 2025 was the year of the AI employee. 2026 is shaping up to be the year of the AI company. The race to own the "AI as your coworker" market is pulling in Big Tech and open-source builders alike — but they're solving different problems. Microsoft, AnthropicDealroom has a profile for this one. Try Dealroom →, and Nvidia are chasing enterprise contracts. Projects like Paperclip are asking: what if you skip the vendor entirely and build the company yourself?
The deeper trend is a shift from single-agent tooling to multi-agent orchestration. Individual AI tools are useful; coordinated AI agents are a different category. If frameworks like Paperclip mature, managing AI agents could become as routine as project management software is today.
Sources:
GitHub
UCStrategies
eWeek
Flowtivity
VibeSparking
ScriptByAI
B.S.
