Paperclip: how to orchestrate an AI-agent company (instead of stacking prompts)

In 2026, the real problem for teams using Claude Code, Codex, or Cursor is no longer access to AI. The problem is orchestration.

You can have 10 capable agents and still lose control: scattered context, runaway costs, duplicated work, and blurry accountability.

That is exactly what Paperclip addresses: moving from "I have bots" to "I run an AI-agent company".

Official project link: paperclipai/paperclip on GitHub

---

Why Paperclip matters now

Paperclip positions itself as an open-source orchestration layer for zero-human companies. In plain terms: if an agent is an employee, Paperclip is the operating system for the company.

This positioning is powerful because it solves a gap most tools ignore:

• Single agents can execute tasks.
• Multi-agent organizations need delegation rules, budgets, priorities, and auditability.

Paperclip provides that missing layer.

---

Our evaluation criteria (AI French Touch framework)

We assess Paperclip with 5 criteria:

1. Business alignment: is every task tied to an economic goal?
2. Risk control: budgets, governance, rollback, execution traces.
3. Operational scalability: can it coordinate 5, 20, then 50 agents?
4. Time to value: how quickly can you run useful workflows?
5. Stack portability: can it integrate heterogeneous agent runtimes?

Against these criteria, Paperclip is one of the strongest frameworks available right now.

---

What Paperclip does better

1) Explicit org chart and roles

Instead of one overloaded "super-agent," you structure hierarchy: CEO bot, CTO bot, marketing, sales, support.

Each agent has a role, manager, scope, and budget.

2) Heartbeats and continuous execution

Paperclip is not a chatbot. Agents run on heartbeats, resume context, and move work forward without constant babysitting.

3) Governance and audit trail

Every conversation is linked to a ticket, every decision is traceable, every action can be explained.

4) Native cost control

Budgets can be assigned per agent. When limits are reached, execution stops.

That is critical to avoid runaway loops and surprise token burn.

---

Practical quickstart

Official quickstart is intentionally simple:

npx paperclipai onboard --yes

Then switch to private network mode if needed, connect your preferred agents, and start with one short-cycle ROI workflow.

---

Where it fits in a B2B sales machine

A. Orchestrated outbound

• Research agent enriches target accounts.
• Messaging agent drafts sequences.
• Ops agent updates CRM and KPI dashboards.

Paperclip handles coordination, priority, and budget governance.

B. Multi-format content engine

• Editorial agent plans SEO angles.
• Repurposing agent adapts into newsletter/post/video.
• QA agent enforces brand consistency.

C. Support and customer success

• Triage agent classifies requests.
• Resolution agent prepares actions.
• Escalation agent routes critical cases to humans.

---

Quick comparison table

Criterion	"Scattered bots" setup	Paperclip
Multi-agent coordination	Weak, manual	Native (org chart + tickets)
Strategic visibility	Fragmented	Goal-aligned
AI cost control	Often missing	Per-agent budgets
Decision traceability	Partial	Full audit trail
24/7 scalability	Fragile	Heartbeats + persistent state

---

Limits to understand

• The "company" mental model has a learning curve.
• Agent roles must be designed clearly.
• It does not replace strategy; it enforces strategy execution.

If you only need a one-off assistant, it may be overkill. If you already run multiple agents, this is the right next layer.

---

Conclusion

Paperclip does not sell AI magic. It delivers operational discipline: goals, hierarchy, budgets, governance.

In B2B environments, that discipline is often the difference between "testing bots" and "building a reliable growth engine".

To explore the source directly, start here: paperclipai/paperclip.