The agent dashboard nobody asked for — and why social teams should still pay attention
If you run social for a living, you already live inside a stack of dashboards that all claim to show you “what’s happening.” Buffer shows you the queue. Metricool shows you the numbers. Later shows you the calendar. None of them show you the thing you actually want at 11pm on a Tuesday: who needs me right now, and why? That’s the gap Cubicle is poking at — not for social scheduling, but for the growing army of AI agents that many creators and small teams are quietly running in the background. The product itself is niche. The pattern it names is not. If you’ve ever lost twenty minutes scrolling Slack trying to figure out which automation broke, or which client draft is waiting on your approval, this launch is worth ten minutes of your attention — even if you never install it.
What Cubicle actually is (in plain English)
Cubicle is a read-only visual dashboard that turns your running AI agents into a tiny pixel-art office. Each agent — a Claude Code session, a Codex or Gemini CLI session, or a “Paperclip employee” — gets a desk. Working agents type with their current task floating above their heads. When one needs your approval, it raises its hand and can ping your desktop. When it finishes, it walks off to the lounge. Errors turn the screen red. The maker, Çağlar Utku GÜLER, frames it as solving a personal problem: he runs a small company of AI agents and kept switching between terminals and dashboards to answer one question — who needs me right now?
The mechanics matter more than the aesthetic. From the maker’s launch post, Cubicle is:
- Read-only by design. It cannot approve, run, or change anything in your agent system. The only optional write is your own Telegram reply posted as a comment.
- Zero dependencies. One Node file and one HTML page. You run
npx @caglarutkuguler/cubicleand you’re in. - Flexible on where it lives. Standalone page, a TV in kiosk mode, or a Paperclip plugin with a dashboard widget.
- Themed. Seven skins at launch — pixel, holding HQ, plaza, warehouse, factory, space, military — plus photo heads cropped from any picture and your logo on the wall.
- Replayable. Record a day and play it back in minutes, or export a 30-second video.
- Followable. Click an agent and the office zooms in, showing its last steps and how long each took.
It’s MIT-licensed, themes are one file each, and there are open “good first issue” tickets for new themes (hospital, newsroom, kitchen, classroom). There’s a live demo you can try with no install.
Why a social media operator should care about a pixel office
Here’s the honest framing: Cubicle is not a social media tool. It doesn’t schedule posts, doesn’t pull analytics, doesn’t touch your Instagram or TikTok accounts. If you’re a solo creator who publishes from your phone, you can stop reading now — this isn’t for you.
But if you’re the kind of operator who has started wiring AI into your content pipeline — using Claude or ChatGPT to draft captions, CapCut or Descript to cut clips, Zapier or Make to move assets between tools — you’ve probably already felt the exact pain Cubicle is describing. You have three browser tabs, a terminal, a Notion doc, and a Slack channel, and none of them agree on what’s actually running. The mental overhead of checking is often worse than the work itself.
I’ve run this experiment myself. Last quarter I set up a small agent chain to repurpose long-form YouTube scripts into LinkedIn posts, X threads, and Threads captions. It worked — until it didn’t, and I spent an afternoon trying to figure out which step had silently stalled. A read-only visual board wouldn’t have fixed the stall. It would have told me where to look in about four seconds instead of forty minutes.
The comparison set (and why it’s thinner than it looks)
The obvious comparison is to observability tools like Datadog or Grafana, which do exactly this for infrastructure — but they’re built for engineers, priced for companies, and require you to instrument everything. Cubicle’s pitch is the opposite: no instrumentation, no config, one npx command.
The closer comparison is to the emerging category of AI-agent dashboards — things like LangSmith, AgentOps, or the various “agent observability” tabs inside Vercel and Replit. Those are deeper, more capable, and more expensive. Cubicle is not trying to compete with them. It’s trying to be the glanceable layer — the thing you leave open on a second monitor while you do other work.
And that’s the interesting move: it’s not selling power, it’s selling ambient awareness. The same instinct that makes a Notion board feel better than a spreadsheet, or a Linear inbox feel better than a Jira queue. The information is identical. The relationship to the information changes.
What social teams can steal from this
Even if you never install Cubicle, four ideas here are worth porting into your own workflow.
1. Separate “working” from “waiting on you”
Most social tools collapse these into one bucket. Your Buffer queue shows scheduled posts, drafts, and failed posts in roughly the same visual language. Cubicle’s distinction — typing = working, raised hand = blocked on a human, lounge = done, red screen = error — is a taxonomy worth copying into your own content ops. In my experience, the single biggest time sink in a multi-platform publishing workflow isn’t creating content, it’s re-triaging content you already thought you’d triaged. If your dashboard doesn’t visually separate “needs a decision” from “in progress,” you’re re-reading everything every time you open it.
2. Read-only is a feature, not a limitation
The maker is explicit that Cubicle only reads. It cannot approve, run, or stop anything. My take: this is the smartest design choice in the whole product. The moment a dashboard can act, it becomes a place you have to trust, audit, and secure. Read-only dashboards are psychologically cheaper to leave open — on a shared TV, on a second monitor, on your phone in a meeting. If you’re building internal content dashboards for a client or a team, steal this constraint. Show state, don’t take action.
3. Notification granularity is the real product
Buried in the comments is the most operationally useful detail: the maker added Telegram controls for /mute, /unmute, /muted, /questions, /errors, and /all. That’s a level of notification hygiene most social tools still don’t offer. Compare that to the average Hootsuite or Later alert setup, where you either get everything or you turn notifications off entirely. The middle ground — “ping me only when an agent is blocked, never when it finishes” — is exactly what a busy operator needs, and almost nobody ships it well.
4. Replay is underrated for post-mortems
The ability to record a day and play it back in minutes, or export a 30-second video, sounds like a gimmick until you imagine using it for a client recap. “Here’s everything that ran on your account yesterday, in 30 seconds.” I’d bet a version of this becomes standard in social analytics tools within a year. Metricool and Sprout Social already lean hard on report exports; a narrative replay is the natural next step.
### Why TikTok creators should care more than LinkedIn ones
This is a niche observation, but it holds. TikTok and short-form video workflows are the most parallel content pipelines in social — multiple clips in flight, multiple captions, multiple platform variants, all at once. That’s exactly the shape of work where an ambient “who needs me” board pays off. LinkedIn workflows, by contrast, tend to be more linear: one post, one draft, one publish. The overhead of a visual dashboard is harder to justify when your pipeline is a single thread. If you’re a short-form operator running five or more clips a day across YouTube Shorts, Reels, and TikTok, the Cubicle pattern is closer to your reality than it looks.
### Where the math breaks
Two honest caveats.
First, the tool is genuinely narrow. It supports Paperclip, Claude Code, Codex, Gemini CLI, and “any tool that writes a small JSON file.” If your agent stack is n8n, Zapier, or a custom OpenAI assistant, you’re either writing the JSON yourself or you’re not in the club. That’s not a flaw — it’s a scope choice — but it means the addressable audience is small and technical.
Second, the aesthetic is doing a lot of work. Pixel offices are charming on day one. On day thirty, when you’re trying to find a failed run at 6pm on a Friday, charm is not what you want. The maker clearly knows this — hence the --redact flag for shared TVs and the settings menu to turn off animation and cards. But I’d want to see how it holds up as a daily driver, not a demo. The live demo is worth ten minutes; whether it survives week three is the real test.
Who this is NOT for
- Solo creators who publish from their phone and don’t run any AI agents.
- Teams already standardized on a heavy observability stack (Datadog, LangSmith) — Cubicle is a downgrade in capability, an upgrade in glanceability, and only you can decide which matters more.
- Anyone who needs the dashboard to do something. It won’t. By design.
What I’d watch / test next
Three concrete things I’d do this week if this pattern interests you:
- Open the live demo and just sit with it for five minutes. Ask yourself whether the shape of the information — working / waiting / done / error — maps onto your own content pipeline. If it does, steal the taxonomy even if you don’t install the tool.
- Audit your notification settings in Buffer, Hootsuite, Later, or whatever you use. Are you getting pinged on completions or only on blockers? If it’s the former, you’re training yourself to ignore alerts. Fix that before you add another dashboard.
- If you do run agents, try the
npx @caglarutkuguler/cubicleinstall and leave it open on a second monitor for a week. Not because you need it — because the experiment will tell you something about how you actually work. My bet: you’ll learn that most of your “checking” time was never about information. It was about anxiety. A read-only board doesn’t fix the anxiety, but it does give it somewhere to land.
