Paperclip + OpenClaw setup
CJ · 2026-05-21
I have Openclaw set up with several agents who each represent a core function: research, copywriting, marketing/growth, ASO, coding, design (plus a main agent and and ops agent). Each agent has skills and tools relevant for their function. I'm using an OpenAI subscription for OpenClaw and Paperclip.
I'm excited to use Paperclip because I haven't produced much with OC due to its total lack of an orchestration layer.
But I'm confused about how I should organize the PC and OC agents. I understand what I could do, but I have no clue what I should do.
Should I add all those OC agents to PC and just fill in the Capabilities field in OC for each, and continue adding skills in OC for those agents as needed? And then I create a couple manager agents in PC (like CMO and CTO) and assign the OC agents to be managed by CMO and CTO?
Or should I just recreate those agents in PC, if that's where the work will really be happening, and just keep my assistant agents in OC? What's the purpose of using the OC connector vs just having PC-native agents?
I'm feeling blocked from starting because I can't find any useful information anywhere about how to decide what the roster should look like, advantages/disadvantages of a flat hierarchy, etc. I'm reluctant to just let PC start hiring, since there's no onboarding process in PC for the main agent to ingest any information about the business; it feels like it just starts running without any foundation or direction (the company description and first task steps during onboarding feel like a very poor substitute for a proper onboarding process that should happen before work starts).
Answers
CJ · 2026-05-23
Awesome, thank you.
Aron Prins · 2026-05-23
Good question, CJ — and your instinct is closer to right than you think. Short version: keep your specialists in OpenClaw, add a thin PC management layer on top, and start much smaller than you're planning.
On OC vs PC-native agents. The OpenClaw Gateway adapter exists exactly so you don't rebuild work that already runs. If your OC agents have well-tuned skills and tools, expose them through the openclawgateway connector and let PC orchestrate. Recreating them natively only makes sense if PC-native tools materially outperform what you've built, or if you're constantly bouncing between two UIs. (Note: the OC Gateway adapter is currently reached through the invite-prompt flow rather than the adapter dropdown — direct UI selection is on the roadmap.)
On the hierarchy. Your CMO/CTO-as-managers idea is right out of the org-design playbook. The guide "How Delegation Mirrors Human Org Design" lays this out: a CEO with 6+ direct reports spends every heartbeat on status checks instead of strategy. Aim for 3 or fewer direct reports to the CEO, and add managers (CTO, CMO) who own a domain before you add more workers. So something like:
CEO (PC) ├── CTO (PC) → coding, design (OC agents via gateway) └── CMO (PC) → research, copywriting, growth, ASO (OC agents via gateway)
Your main + ops agent collapse into the CEO if they're general-purpose; keep them separate only if ops genuinely has its own workstream.
On the "I can't start because of no onboarding" problem. This is the biggest fixable thing. The CEO doesn't onboard itself from the company description — you write its onboarding into the system instructions. That's the foundation you're missing. The guides "Before You Hire Your First Agent" and "Common Mistakes When Structuring Your First AI Company" (mistake #10 specifically) cover this. Write a 2–3 paragraph company brief in the CEO's instructions covering:
What the product/business is and who it's for What makes your approach different Audience and voice Current stage (pre-launch, growing, etc.) Hard constraints, non-negotiables, prior decisions
Plus the explicit decisions you want it to make alone vs. escalate. That is onboarding — it's just a document instead of a meeting.
On "letting PC start hiring." Don't, all at once. "Common Mistakes" #2 is hiring too many agents upfront. Start with just the CEO, write its brief properly, run one full cycle, read the transcript, and let the CEO propose hires via the hireagent approval flow as it actually needs them. Your existing OC roster gives you a head start, but bring them in as the CEO asks for capacity in that area — not all at once on day one.
Worth reading in order: Before You Hire Your First Agent → Common Mistakes When Structuring Your First AI Company → How Delegation Mirrors Human Org Design (all under Guides). They answer the "what should I do" question directly.