Run 10 AI agents on one screen by assigning each a role, a token limit, and a schedule inside one dashboard, then approving deliverables as they arrive. That is the workflow Julian Goldie's October 2026 Agent OS demo describes, using an agent org chart plus a desktop chat client wired together.
How to run 10 AI agents on one screen
To run 10 AI agents on one screen, you connect each agent to a single orchestrating dashboard, give every agent a role and a token limit, and manage the whole group from one org chart. That is the core workflow demonstrated in the video "Paperclip + Hermes Just Changes Everything!" published on 10 October 2026 by Julian Goldie of Goldie Agency.
The demonstration, which Goldie presents through his digital avatar, combines two tools he names Paperclip and Hermes. Paperclip, as shown in the video, organizes agents into a company-style hierarchy; Hermes is the chat client he uses for conversation, voice, and media work. The transcript is the primary source for every capability described below, so treat the specifics as the presenter's first-hand demo rather than independently verified benchmarks.
The video frames the result as "mission control": voice chat, a media studio, past sessions, MCP integrations, and a live control room in one view. The presenter says the dashboard itself was built by the agents running in the background, with what he describes as no human developers involved in that build.
Why scattered tabs fail as an agent workflow
Scattered tabs fail because each chat lives in its own window with its own context, so nothing is visible in one place. The video describes the common pattern as one tab for Claude, Anthropic AI assistant, one for other tools, and around twenty windows in total, with work copied and pasted between them.
Goldie's argument is that a pile of chats is not a team. In the tab-based workflow, he says, you re-explain the task every session, and a laptop restart wipes the working state. The Agent OS approach instead keeps agents, their outputs, and their history in one interface so you can see who is working and on what, live.
The tools he names as connectable include Claude Code, Anthropic agentic coding tool that runs in the terminal, available at its GitHub repository, plus OpenAI's Codex CLI, which is open source at github.com/OpenAI/codex. The claim that all of these plug into one org chart comes from the presenter, not from those projects' own documentation.
Inside the agent org chart: roles, bosses, token limits
Every agent in the demo is set up like an employee with a role, a title, a boss, and a token limit. In the video's org chart, an agent named Ada sits at the top as CEO, with other leaders below her and teams beneath them: 10 agents across 3 departments.
The framework the presenter describes has five layers, starting with you as the founder. You set the mission and approve major moves; everything else is delegated down the hierarchy. Hiring is a click: choose a role, and the agent joins the chart.
Two controls matter most in day-to-day use:
- Token limits. A token is a small piece of AI work. When an agent hits its limit, it stops, which the presenter presents as a guardrail against surprise usage.
- Schedules, called heartbeats. You choose when each agent works, pause it for a break, resume it, and review its latest runs with timestamps.
The comparison between the two workflows, as the video presents it, looks like this:
| Dimension | Many-tabs workflow | Agent OS team workflow |
|---|---|---|
| Context | Per-chat, lost between sessions | Shared dashboard with sessions and history |
| Oversight | One window per tool | Live view of all agents, filterable |
| Cost control | Ad hoc per chat | Per-agent token limits |
| Work style | Continuous prompting | Set a goal, review deliverables |
Voice assistant, Claude Code, and the studio
The demo bundles several capabilities into the same dashboard. The flagship feature is a voice assistant the presenter calls Hermes Jarvis: it wakes on a spoken wake word, responds by voice, previews what it built, and has adjustable voice settings.
Claude Code, which you can install from Anthropic official repository, is shown built into the interface: you speak an instruction, the agent builds, and a live preview renders the result. The integrations rely on MCP, the Model Context Protocol that Anthropic published as an open standard at modelcontextprotocol.io, for connecting external tools and data.
Other panels shown in the video include a studio for generating video, images, and voice; a sessions view listing previously built apps and projects; and support for multiple teams, with the demo showing Goldie Agency, Goldie Labs, and Julian Goldie's personal team side by side. The presenter also says agents can run on local models, meaning you pick the model and the agents use it, though the video does not specify which models or hardware were tested.
Setting up an agent team, step by step
The setup process shown in the video takes four steps from empty dashboard to running team. Each step maps to a control in the interface demonstrated on screen.
- Hire your first agent. Click hire and pick a role; the agent appears in the org chart.
2. Give the team a goal. You assign a task the way you would to a real team member, then step away instead of prompting continuously.
3. Set token limits and schedules. Cap each agent's spend and choose its heartbeat, the interval at which it works.
4. Review and approve. Watch the live board, check each agent's latest runs, and accept the deliverables it produces.
The presenter's advice is to start small. In his telling, the Agent OS improved through daily iteration, and he reports seeing people with no prior AI experience get results with systems of this kind. Those outcomes are his observation from his community, not a published study.
What the video does and does not prove
The video proves one thing directly: the presenter ran a working demo on camera in October 2026 with 10 agents, 3 departments, and voice control in a single dashboard. Everything else, including the claim that the agents built the dashboard with zero human developers, is a vendor-side claim from the person selling an accompanying paid community called the AI Profit Boardroom, so weigh it accordingly.
A few specifics stay unverified. The video does not name the underlying models beyond general references to Claude Hermes, does not publish token usage numbers, and does not show independent testing of the "runs on local models" feature. The pitch copy in the description, including a shared-memory Agent OS zip file and an Obsidian knowledge-base setup, where Obsidian is a note-taking app built on local Markdown files, describes what the paid community ships rather than a publicly documented product.
If you want to evaluate the approach yourself, the safe path is to reproduce the structure with tools you already control: pick an orchestrator, connect one coding agent such as Claude Code through MCP, set per-agent budgets, and measure deliverable quality against your current tab-based workflow before committing.
FAQ
- How many agents does the demo run at once? The October 2026 video shows 10 AI agents organized into 3 departments, with an agent named Ada at the top of the org chart and the presenter approving work from a single dashboard.
- Do you need to prompt the agents all day? No, according to the demo. You assign a goal, the agents work on their scheduled heartbeat, and each task ends with a deliverable you review and approve.
- What happens when an agent hits its token limit? The agent stops. The presenter describes the token limit as a per-agent budget so costs stay predictable instead of accumulating silently.
- Can the agents use different AI tools? The video says Paperclip connects to multiple agents, including Claude Code and Hermes, all shown in one org chart, and that agents can run on local models you select.
- Is the Agent OS a product you can buy separately? The video distributes it as a ready-to-install zip file inside the presenter's paid AI Profit Boardroom community, along with tutorials and coaching calls; it is not presented as a standalone public download.
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits