Paperclip: The Open-Source Platform Turning AI Agents into an Actual Company
Source: Kristopher Dunham, Paperclip: The Open-Source Platform Turning AI Agents into an Actual Company (Medium, 2026-04-06)
Key takeaways
- Paperclip is an open-source platform structuring AI agents as company roles (CEO, CTO, engineers, PMs).
- Competitive with — but distinct from — Anthropic Managed Agents.
- Worth knowing about even if you don’t adopt: the design philosophy (“agents should inherit org-chart structure, not graph nodes”) is an interesting data point for ACT architecture.
- Open-source = you can read the code, which is more than you can say for Managed Agents or Copilot Cowork.
The pitch
Most multi-agent frameworks represent agents as nodes in a graph (LangGraph, AutoGen) or as roles in a pipeline (CrewAI). Paperclip structures them as a company — with departments, reporting lines, role definitions, and cross-team coordination.
The argument: humans already know how companies work. Operating an AI workforce using company semantics is less mental overhead than operating a graph.
Why it matters (even if you don’t adopt)
- ACT comparison: Paperclip’s “agents-as-company” is a third architecture option alongside Rezvani’s orchestrator pattern and Anthropic’s Managed Agents. Worth comparing when making architectural decisions for ACT.
- Open source: read the code. Paperclip’s repo documents design decisions in-source in a way Managed Agents can’t (closed) and CrewAI doesn’t explicitly.
- Domain fit for regulated verticals: company-structure semantics map onto compliance hierarchies (CRO, CISO, SOX auditor roles) more naturally than graph semantics.
When to look at Paperclip
- You’re evaluating multi-agent orchestration options for ACT or a similar platform
- You want an OSS alternative to compare against Anthropic Managed Agents
- You’re designing for regulated customers and want a structure that maps onto corporate hierarchies
When to skip it
- You’re already committed to a multi-agent framework that works
- Your use cases are single-agent or light-orchestration
- Your team doesn’t have the bandwidth to evaluate yet-another-framework
Related Playbook pages
- Multi-Model Orchestration — full orchestration comparison
- Anthropic Managed Agents launch — the first-party comparison point
- BMad Autonomous Development — another concrete pattern worth comparing