§ 01 · The Hook
The bait, then the rug-pull.
For months the multi-agent management problem was a DIY sport: Vibe Kanban, TMux scripts, homemade dashboards. Then Anthropic shipped something native — and Simon Scrapes spent five minutes showing exactly how to use it.
What the video promised.
stated at 00:49 "Let me show you how to get the most out of it and how it pairs with your own Agentic operating system." delivered at 04:55
Where the time goes.
00:00 – 00:49
01 · The problem + the reveal
Agents are powerful but multi-agent management is chaos — five terminal windows, community workarounds, dashboard sprawl. Anthropic shipped Agent View natively inside Claude Code.
00:49 – 01:07
02 · How to enable it
Must be on Claude Code v2.10.139+. Run 'claude --version' to check, 'claude --update' to upgrade, then 'claude agents' to enter the summary view.
01:07 – 02:25
03 · Migrating sessions + basic navigation
Use /bg to background existing terminal sessions — they appear instantly in Agent View. Sorted by status by default. Ctrl+S for repo sort, Shift+Up/Down to reorder, Ctrl+T to pin.
02:25 – 03:44
04 · Current limitations
Repo sort only works at the parent folder level — no subfolder grouping for multi-client setups yet. Simon demonstrates the gap and flags it as a future request.
03:44 – 04:55
05 · Deep-diving sessions + approvals
Jump into any session for full context view. Spacebar opens a quick-reply interface. Typing 'approve' attempted batch approval — partially worked, still needed manual jump-in.
04:55 – 05:35
06 · Pairing with the Agentic OS
The Agentic OS is just a folder structure that injects context (brand, client, scheduled jobs) at the right time. Agent View is the control plane on top. Each session benefits from the context automatically.
Visual structure at a glance.
Named ideas worth stealing.
04:55
model
The Agentic OS
A folder structure that injects context (brand, client, scheduled jobs) automatically into Claude Code sessions via CLAUDE.md files at the right directory levels. Agent View becomes the control plane on top.
Steal for JoeFlow Sessions panel architecture — each project folder = context scope, Agent View = cockpit
02:25
list
Agent View keyboard shortcuts
- claude agents — enter the TUI
- /bg — background current session into dashboard
- Ctrl+S — toggle sort (status vs repo)
- Shift+Up/Down — manually reorder sessions
- Ctrl+T — pin a session
- Spacebar — open quick-reply from summary view
- Left/Right arrows — navigate in/out of session detail
Steal for Any tutorial about Claude Code multi-agent setups
Lines you could clip.
00:42
"It was dashboard crazy. So it's honestly about the time that Anthropic stepped in and shipped something native."
05:01
"The AgenTek OS is just a folder structure at the end of the day that injects context at the right time."
02:40
"You can have loads of agents operating at the same time, see this summary level view, and then actually pin some or jump into the detail if you want to."
How they spent the runtime.
Hook length49s
Info densityhigh
Filler5%
Things they pointed at.
00:00toolVibe Kanban
00:00toolTMux
How they asked for the click.
05:19
subscribe
"Make sure you subscribe if you haven't already. See you in the next one."
Low-key subscribe ask at the very end after the main content is complete. No hard pitch.
Word for word.
HOOK opening / re-engagementCTA the pitch metaphor analogy
00:00HOOKAgents have gotten really good, and we're no longer sitting in one ChatGPT window babysitting every step and asking it to just redo everything. They're now actually executing on a brief and getting us to 90% from a good prompt. But this has created a brand new problem for us. How do we actually manage multiple agents running at the same time without five terminal windows open and juggling between them? So we had Vibe Kanban who built a Kanban style UI,
00:25HOOKTMux who had an unofficial answer for the terminal crowd, and I even built my own command center to manage business goals. And loads of others did exactly the same. It was . So it's honestly about the time that Anthropic stepped in and shipped something native for us to all use. And that's what we're looking at today, the new agent view. So it's a proper UI for managing multiple ClawCode agents built right into ClawCode.
00:49So let me show you how to get the most out of it and how it pairs with your own Agencik operating system. So first thing you need to do, because it's in research preview, is make sure that you're on the latest update for Claude. So you can hit Claude dash dash version. It will tell you exactly which version you're on, and you need to be on 2.10.139 or higher. So we're gonna go Claude dash dash update in our terminal, and it's gonna effectively make sure you're up to date. We're then gonna just run Claude agents, and you're gonna pop into the summary view, and you're gonna gonna become really familiar with this view because it's the high level view of how we see all of our agents organized by session. Now the first thing you're probably going to want to do is migrate your existing sessions into this dashboard view. So if you've got any existing sessions
01:32and you start the agent view, this will appear blank. So we'll go to an existing session, and what we can do is just hit slash b g, continue this session in the background, and free the terminal. It's gonna background that task. And then when we go back to the agent view, we've got another session that's appeared inside our agent view. We'll do the same with this session here. It's gonna background it, and you see these these sessions effectively close.
01:54So what we can do now is actually close down these two terminals. They've been backgrounded, and they've been put into the agent view across here. So by default, it's sorted by status. It makes total sense from a UX perspective. We can describe a new task for a new session down here. So we're spinning up a new agent or a new instance when we describe a task down here. So draft a LinkedIn carousel
02:13based on my last YouTube video. I'm gonna hit that, and you can now see that that session is working. So it's sorted by status by default, but this is where it gets really powerful because if you're working across multiple repositories, you can actually hit control s and sort it by repo instead or folder structure. So here we're working all tasks in the same repo, and you've got this sorting here. And what we can actually do is go up and down the list here. We can actually use the shift up and shift down to move any session in any order that we want, and we can even pin them with control t if you wanted to. So you can have loads of agents operating at the same time, see this summary level view, and then actually pin some or jump into the detail if you want to. Now right now, this sorting by repo does not go down to the subfolder level, which is something I'm hoping they're gonna add in the future. The way that we organize our agentic operating system is that we have a client's folder, and we jump into that individual client's context. So say I was operating within that client's folder. You can see up here client slash Acme Corp. Tell me about the brand voice for this client. So we're operating within that client folder, but when we actually go backwards
03:16into the agent view, it still appears under the same sorting at the parent folder level. What would be brilliant in the future is if we can sort by subfolder levels so that we can actually operate within multiple clients and just see all the tasks for all of our different clients in separate aggregated groups down here. So if we jump back to the status view, on those that need input, we can see the names of the different sessions and flick through those. We can see the last output that Claude code has given us and also when we last discussed something with this conversation
03:44or how long it's been waiting for our input, for example. Now we can, of course, jump into the detail of any of these conversations. So if we wanted to jump into the Metascale system creator, we now get a recap at the bottom. We get the ability to control this session specifically from the remote control view, and we can reply directly in the conversation but still see all the full context as if it was a Clawd code terminal session. So we just need to hit back on our keyboard or left on our keyboard to jump back and right to jump into that session. Now what's more useful is you can actually talk to those sessions directly from this interface. So if we hit space bar and we go up and down, we'd the latest status for all these different conversations, and we can reply directly in here. Let's see if we can actually
04:24approve from this window too. So if a couple of them are awaiting approvals, let's see if actually by typing approve, we can push through the approval at this point. Now there is one limitation here. It's terminal only at the moment, so we don't have access to the desktop app when doing this. But I'm sure soon in the desktop app, they'll have a similar interface to interact with multiple sessions. It didn't seem to actually give the approval in that session, so we needed to jump in and actually approve it manually. Now the big question is you set up your Agencik operating or your Claude operating system.
04:55How do we actually use this in conjunction with that? And it's actually pretty simple because the AgenTek OS is just a folder structure at the end of the day that injects context at the right time. So it's just pulling in your brand context, your client context, your scheduled jobs at the right time. So if you've got your operating system architecture already built, then each conversation is gonna benefit from the context.
05:17CTAAnd now you can just operate multiple conversations at one time without flicking through multiple terminals to try and work out which ones need input and which ones don't. So go and check it out for yourself and see what you think. I'm curious whether this will stop you using other UIs, so let me know below and make sure you subscribe if you haven't already. See you in the next one.
— full transcript
§ 05 · For Joe
The two-layer pattern.
Steal this architecture
Agent View is the cockpit; the Agentic OS folder structure is the engine — and you can build the engine first, then drop the cockpit on top later.
- Map your projects as folders with CLAUDE.md context files at each level — this IS the Agentic OS.
- Use /bg to hand off any running session to the dashboard without losing state.
- The subfolder-sort gap is a real product hole — position JoeFlow's Sessions panel as the fix (client-level grouping out of the box).
- Simon's 'limitation-first' honesty pattern (show what doesn't work) is a trust-builder worth copying in your own tutorials.
- The format: Problem → Reveal → Enable → Demo → Limitation → Integration — all in under 6 minutes. Time it for your next tool tutorial.
§ 05 · For You
How to actually run multiple AI agents without losing your mind.
For the multi-agent curious
Claude Code's new Agent View is the closest thing to air-traffic control for your AI agents — one screen, every session, no terminal juggling.
- Update Claude Code to v2.10.139+ and run 'claude agents' to try it immediately.
- Use /bg in any running Claude session to send it to the dashboard without stopping it.
- Ctrl+S sorts your sessions by project folder — useful when you're running agents on different codebases.
- You can reply to any agent directly from the dashboard without opening a new terminal.
- It's terminal-only for now — the desktop app version is coming.
