Paperclip Promises a Company of AI Agents but Drowns in Bureaucracy During Testing

Paperclip offers the ability to manage not just a single AI assistant, but an entire "company" of agents with roles, budgets, and internal hierarchy. The Habr author tested the idea on a real product task and got a striking failure: over two days, a team of eight virtual "employees" burned through tokens, created noise, and accomplished almost nothing.

How

Paperclip Works Paperclip is an open-source project that, according to its author, gathered 24 thousand GitHub stars in its first 12 days and reached 30.6 thousand by the time of publication. The idea is simple yet ambitious: stop working with AI as a set of separate chats and transform interaction into a company model. Instead of a single assistant, a user gets a CEO, CTO, developers, a designer, and other "employees" between whom tasks, budgets, and responsibilities can be divided.

"If OpenClaw is an employee, then Paperclip is a company".

• CEO, CTO, engineers, and designer roles for different types of tasks token limits and disabling agents when budget is exhausted heartbeat mechanism that wakes agents on schedule rather than keeping them constantly active connection of each task to an overall goal so agents understand the work context support for different tools — from Claude Code and Codex to Bash and HTTP agents On paper, this looks very powerful. Paperclip attempts to solve a real pain point for developers who have dozens of agent sessions open simultaneously, lose context, and see token expenses spiral out of control. An ecosystem is already forming around the project: cloud deployments, active community, and the idea of a marketplace for ready-made "AI companies" for typical scenarios — from marketing to development. This explains why interest in the project grew so quickly.

Why

Everything Stalled To test the idea, the author gave Paperclip a fairly typical task: finish an existing product, add a couple of endpoints, integrate with an external API, and bring the UI up to the already-drawn design. Then the system behaved all too humanly. The CEO distributed tasks and disappeared to check metrics.

DevOps got caught up in automation and generated a pile of CI/CD configs. QA refused to work with this format. Backend saw the scope of the specification, irritably backed away.

The designer limited themselves to a remark like "make the buttons like in other places". The team lead periodically woke up, issued strategic directives, and went back to sleep. By the second day, things hadn't improved.

Backend returned, saw new clarifications and a long thread of messages, then went into deep timeout. Frontend sat waiting for an API that never came. The experiment's result was almost satirical: eight agents over two days spent 79 thousand tokens, wrote not a single useful line of code, broke a job card, and drove the author to revert the changes.

After that, a solo contractor in the form of Cursor closed the same task in roughly an hour and a half.

Where

Is the Real Value The author themselves makes an important caveat: the problem isn't that Paperclip is useless. Quite the opposite — it hit a very precise market pain point. When a developer has Claude Code, Cursor, Codex, and other agents running simultaneously, chaos quickly emerges: unclear who's doing what, where discussion history is, what context was lost, and how much money has already been spent.

Paperclip offers a management layer on top of this zoo — with tickets, auditing, state recovery, and budget control. But the experiment also reveals the main risk. As soon as interaction is structured around a regular company model, alongside useful division of labor come bureaucracy, coordination losses, and dependence on the quality of each individual agent.

Paperclip doesn't make models smarter and doesn't fix bad decisions — it merely organizes them into a structure. Therefore, the most sensible human role in such a system today is not a micromanager and not an "employee alongside the bots," but a board of directors: set goals, allocate budgets, watch metrics, and intervene only where autonomy truly breaks down.

What

This Means Paperclip definitely hit on a future market problem: managing multiple agents is already harder than working with one. But the article illustrates well the boundary of current capabilities: coordination tools are needed right now, but copying a human company with its hierarchy and rituals for autonomous agents is premature.