I Watched My AI Agents Work In A Pixel Office

My entire agent OS runs on Hermes, and until this week I had never actually seen it work.

I saw output. I saw logs. I saw a cursor blinking while something happened somewhere.

Then I installed the Hermes Pixel Office, and my agents turned into little pixel people walking into a room, sitting at desks, and getting on with it.

It sounds like a toy. It changed how I run the whole thing.

What it is, quickly

The Hermes Pixel Office comes from Teknium, one of the people behind Nous Research — the same team that builds Hermes Agent.

It is free and open source. Nothing to sign up for, nothing to unlock.

Every agent session becomes a pixel character at a desk in a shared office you can watch live.

The problem I had stopped noticing

Hermes agents work in the background.

They run tools, they spin up helper agents, they do real work — and you see none of it.

You stare at scrolling text hoping nothing is quietly stuck.

I had been doing that for months and had genuinely stopped noticing how bad it was.

You only see it once the alternative exists.

What I actually saw

The characters walk in through the door, sit down and get to work.

When they finish, they stand up and walk back out.

Helper agents — the ones a bigger agent spins up for a side task — arrive with a gold collar and a name tag showing their job.

So within a second of looking I know who is the main worker and who is a helper it pulled in.

And they act out what they are doing.

Writing a file: the character is typing.

Reading something: a little book is open.

Browsing the web: it is browsing.

Running a terminal command: a green monitor flickers.

I have not read a log line in days.

Watching a content series get built

The first real thing I used it for was a content series for the Boardroom.

I kicked off a batch of Hermes agents to draft hooks, scripts and captions.

In the office I watched each helper pop up as a gold-collared character — one typing away, one reading through notes, all working simultaneously.

When one hit a step that needed my approval, a red flag appeared over its head.

I knew the exact second to step in, instead of finding out twenty minutes later that everything had been waiting on me.

The approval flags are the real feature

Sometimes an agent hits a command that could be risky and needs a yes or no.

That character throws up a bright red exclamation mark, and a counter at the top of the screen shows something like "2 waiting".

There is an optional chime as well — off by default — that dings when an agent needs your okay or when a helper finishes.

Turn it on and you do not even need to be looking at the screen.

That single feature removed the thing I hated most about running agent batches: not knowing whether progress had stopped or whether it was just slow.

Old way vs new way

Old way	New way
Read scrolling logs and guess.	Look at a room and know instantly.
Discover a stall an hour later.	See the character stop moving.
Hunt for approval prompts in text.	Red flag over the head, counter at the top.
No idea which agent is which.	Gold collars and name tags on helpers.
Wonder if the batch finished.	Characters walk out when they are done.

Two more things I used it for

Member onboarding. I had agents lay out the welcome messages, first-call prep and resource walkthroughs. I watched each character finish and walk out the door, which told me at a glance the whole flow was complete and nothing was silently stuck.

A lead system. Fire off a group of agents to plan content, hooks and follow-ups, then watch the office fill up. One look tells me who is working, who finished, and who is waiting on me — the whole thing from one small screen instead of ten open tabs.

The myth I had believed

I assumed a live dashboard like this would slow my agents down, or get in the way, or drag on them under the hood.

It does not.

The Pixel Office is a window and nothing more.

It watches what your agents are already doing and draws it out.

It never blocks them, never changes their work, and never adds weight to how they run.

The agents do not even know it is there.

That is what makes it click for me — I am not giving anything up to get this view.

🔥 Want my exact Agent OS with Hermes as the brain?

Inside the AI Profit Boardroom you get the full Agent OS zip ready to install, a 30-day roadmap, Pixel Office and Hermes walkthroughs, and four live coaching calls every week with 4,000+ members.

→ Get access here

Where to run it

Two options, same office.

Open it in your browser as its own page.

Or drop it into VS Code so the office sits next to your code, with a "+ agent" button that starts a fresh Hermes run without leaving the editor.

Every Hermes agent on your machine shares the same office, so several jobs at once all pile into one room.

It also works over a remote connection out of the box, which matters to me because my agent operating system runs on a server.

Four things I would tell you before you install it

Try demo mode first. It fills the office with pretend agents so you learn what every character means before wiring up your real setup.

Use the VS Code panel if you live in VS Code. Same office, less context switching.

Keep your safety rules on. It is a window, not a controller. The red flag tells you when to look; you still decide yes or no.

Do not expect it to fix a broken workflow. It shows you the truth faster. That is genuinely valuable, and it is not the same as doing the work.

FAQ

What is the Hermes Pixel Office? A free, open source dashboard from Teknium at Nous Research that shows each Hermes agent as a pixel character working in a shared office.

Does it slow agents down? No. It only draws what your agents already do — it never blocks or changes their work.

How do approval alerts work? A red exclamation mark appears over the agent that needs a yes or no, with a counter at the top and an optional chime.

Can I use it with a remote server? Yes, it works over a remote connection out of the box.

Do helper agents look different? Yes — gold collar and a name tag showing the job they were spun up for.

Why seeing changes behaviour

There is a version of this that sounds like a gimmick, and I understand why.

Pixel characters at desks is a design choice, not a technical breakthrough.

But watch yourself use it for a day and you notice something.

You start kicking off bigger batches, because you are no longer nervous about losing track of them.

You start delegating steps you would previously have done yourself, because you can see whether the agent got stuck.

Visibility does not make agents better. It makes you braver with them.

That is a real productivity change and it has nothing to do with the model.

What it does not fix

It does not make a badly designed workflow work.

If your agents are doing the wrong thing, you will now watch them do the wrong thing in higher definition.

It also is not an audit trail — it is a live read on state, not a record you can go back through.

Keep your logging, keep your safety rules on, and treat the office as the fastest way to answer two questions: is anything stuck, and does anything need me.

Those two questions are most of what supervision actually is.

About Julian

I'm Julian Goldie — AI entrepreneur, SEO expert, and founder of the AI Profit Boardroom (4,000+ members). I help business owners scale with AI agents, automation, and SEO.