Paperclip blog

Braintrust, Arize Phoenix, LangSmith, Galileo, AgentCenter, and Paperclip compared by the record each one keeps: the agent run, or the work around it.

Give an existing AI agent a task, answer its questions, check the result and approve it from the Paperclip board, once someone technical has set it up.

Test whether your campaign needs independent review, separate access or lasting ownership before adding agents. Includes a filled launch handoff.

What AI agents can finish overnight, what should wait for a person, which layer enforces each limit, and what to check in the morning.

When Claude Code and Codex share work, which layer stops a risky command, which resumes a broken session, and which still knows who owns the task?

Compare Agent View, Agent Teams, amux, and Paperclip for managing Claude Code and Codex sessions with ownership, budgets, approvals, and audit trails.

Paperclip, LangGraph, CrewAI, AutoGen, LlamaIndex, and OpenHands all publish MIT-licensed code. What each asks you to run, where model calls still go, and what your team maintains.

Seven tools to organize your AI agents, manage spending, and keep important decisions in your hands.

You choose the apps, accounts, and actions available to each agent.

Learn what an AI agent control plane does, the five layers products govern, and how to evaluate ownership, approvals, budgets, and evidence.

AI agents can handle the coordination that keeps serious work moving, as long as you define the goals and guidelines.

Paperclip v2026.824.0 turns the task page into a conversation and adds in-product Claude and Codex sign-in plus previews over Tailscale HTTPS.

Decisions let an agent ask permission for a specific action. Agents can coordinate across tasks, and one Activity page shows who changed what.

AI agents can produce work faster than you can review it. Learn how to define done, require proof, and turn every quality decision into a better next run.

Paperclip agents can now request an approved secret through an API instead of holding it in the environment for the full run. Every request is logged.

Stop opening issues one by one. The Artifacts page indexes every work product across issues and runs, grouped by task stack, with rich playback.

The v2026.609.0 catch-up: an Artifacts page, inline video, checkbox confirmations, and a bundled wireframe skill make agent output visible and steerable.

Install the bundled wireframe skill, prompt an agent with a screen list, and get a reviewable deck: SVG screens, a flow map, and a published HTML viewer.

Design conversations used to start with a build; now an agent turns a one-line prompt into a reviewable multi-screen wireframe flow.

v2026.707.0 makes team operation concrete: per-user secrets with a dispatch check, one Work Timeline for all agent work, and self-serve sandbox setup.

Secrets can now be scoped to each user: teammates store their own values, and Paperclip checks the human behind a run supplied them before it dispatches.

Ramp for Agents is live. Give a Paperclip agent Ramp's get-started skill and it can incorporate, apply, and stand up your finance stack while you approve what matters.

Generate a one-time onboarding prompt in Paperclip, paste it to your running Hermes agent, and approve its join request — no adapter config by hand.

Hermes is now built in, watchdogs watch long-running tasks for you, and ask mode gets you answers without a workflow. 122 commits from 17 contributors.

Watch a live demo of Paperclip in action — hiring, managing, and orchestrating AI agents as if they were real employees.

I reviewed 14,267 public Paperclip-attributed commits across 884 repos to see what people are building, who keeps coming back, and what that says about the real customer profile.