Paperclip: Open Source Human Control Plane for AI Labor — Dotta Bippa
TL;DR
Paperclip positions itself as the “human control plane for AI labor” — Dotta Bippa frames it less as a magic autobuilder and more as an open-source org chart where humans stay accountable, set goals, hire agents, and inject taste into the work.
The core workflow is surprisingly simple:
npx paperclip-ai onboardand start with a CEO agent — from there, the CEO can hire roles like CTO, coders, marketers, and QA, then break tasks down across the org instead of making you juggle 30 disconnected Claude or Codex tabs.Paperclip’s big pitch is shared organizational context, not just agent access — in his 40,000 GitHub stars example, an agent made a Remotion celebration video in about 5 minutes because it already had the stats dashboard, brand guide, and installed skills inside the system.
Reliability comes from workflows like reviewer and approver gates — Bippa argues that agents often ignore instructions like “test this in the browser,” so Paperclip adds vendor-neutral QA, browser automation, and approval chains on top of models from Claude, Codex, Gemini, Hermes, Pi, and more.
He explicitly warns against spawning giant 130-agent orgs on day one — his advice is to build agent by agent, tune instructions constantly, and reserve expensive frontier models for the highest-leverage roles while using cheaper options through OpenRouter for “good enough” tasks.
The product is very early, but moving fast — Paperclip launched March 4, hit 40,000 GitHub stars within roughly 34 days, crossed 50,000 live during the demo, and has near-term roadmap items like multi-user support, cloud deployment, sandboxing, CEO chat, and “maximizer mode.”
The Breakdown
The pitch: not a zero-human fantasy, but managed AI labor
Dotta Bippa opens with the flashy tagline — “open-source orchestration for zero-human companies” — then immediately refines it into something more grounded: Paperclip is the “human control plane for AI labor.” His point is that you’re still accountable for the work; Paperclip just gives you an org chart of agents you can direct, shape, and manage.
A tour of the company he actually uses to run Paperclip
Instead of a toy demo, he shows the live Paperclip org behind Paperclip itself: a CEO, a CTO, multiple coders, and a marketing team with roles like content strategist and video writer. The notable twist is “bring your own agent” — Claude, Codex, Gemini, Hermes, Pi, OpenClaw, and others can all become employees inside one shared memory and communication system.
The 40,000-star video that took five minutes instead of a week
The most concrete example is a celebratory Remotion video for Paperclip crossing 40,000 GitHub stars. He assigned the task to the CEO, who hired a video writer, installed a Remotion best-practices skill, wrote a plan, got feedback like “cuts should be 2 seconds, not 6,” and then produced an on-brand animation with real stats in about five minutes.
Why this beats one-off prompting in Claude Code
Bippa’s argument is that the magic wasn’t the prompt — it was the built-in context. The agent already knew Paperclip’s branding guide, had access to the existing stats dashboard, and could accumulate feedback into organization-specific skills, so the work didn’t require re-explaining the whole company every single time.
QA, reviewers, and the messy reality of agent reliability
He spends a chunk of time on what actually makes orchestration useful: guardrails. Paperclip can require a QA agent with browser skills to review work before it comes back, and can separate reviewer from approver — because, as he jokes, we’ve all asked an agent to “please test this in the browser” and watched it simply not do it.
Routines, reusable tasks, and a bigger vision than coding
He demos routines for recurring work like summarizing merged PRs, writing release notes, or turning Twitter bookmarks into reports, complete with template variables and skill reuse. But he’s emphatic that Paperclip is not just a coding tool: the real ambition is for marketing, sales, finance, and operations teams to use it as a general AI-labor wrangling system.
Starting a new company and the advice not to overbuild
Bippa creates a fresh company live — an “agent tools directory” business — with a CEO who drafts a hiring plan and requests permission to hire a CTO. His guidance is practical: don’t import some massive default org, start with the agents you actually need, tune them carefully, and mix premium and cheaper models depending on how much intelligence each role really needs.
Early product, fast roadmap, and a live 50,000-star moment
He’s candid that this is “very, very early” software, with gaps like no multi-user support yet, plus experimental workspaces and lots still in progress. Then, mid-roadmap walkthrough, he notices Paperclip has just crossed 50,000 GitHub stars live on screen — a perfect on-brand moment for a product about coordinated AI work growing absurdly fast.