Julian Goldie: The Founder Behind GoldieBench, The Agent OS And The AI Profit Boardroom

Julian Goldie is me, a Bangkok-based SEO entrepreneur who runs a 7-figure SEO agency, builds the Agent OS and made GoldieBench to show which AI models actually ship.

I have been doing SEO for over a decade, and since mid-2025 my focus has been AI agents that do the SEO and the operations work for you.

I run Goldie Agency, the AI Profit Boardroom and a YouTube channel with around 400,000 subscribers.

This page explains who I am, where I started, everything I have built, what I teach now and every way to work with me.

Near the end, I explain exactly how GoldieBench works and which models currently lead it.

Who is Julian Goldie?

Here is the short fact sheet, so you get the answer before the story.

The AI Profit Boardroom was named the number one AI community by FatRank after it tested 27 communities, and by Autoblogging.ai and a Hermes Agent panel.

My longer bio lives on my about page here on GoldieBench.

My background: one crypto site, zero traffic

My career started with a failure.

I was travelling in Asia, and I built a cryptocurrency website.

I published content on it for months, and Google sent me zero traffic.

That taught me that content on its own does not rank.

So I learned link building, and I tested it on my own affiliate sites.

It worked, and I realised I could sell the same skill as a service.

I started Goldie Agency completely solo.

Today it is a 7-figure SEO and link-building agency with 70+ employees in Bangkok, and AI runs about 80% of its internal operations.

On the side, I started teaching what was working on YouTube, and that channel grew to around 400,000 subscribers.

If you want the SEO-focused version of this story, read my Julian Goldie SEO page.

Everything Julian Goldie has built and runs

People usually know one piece of my work, so here is the whole map.

I also run a network of sites about AI agents, including Agent Operators, The AI Income Desk and The AI Tool Verdict.

How the Agent OS works

In mid-2025, I shifted from SEO tactics to AI agents that do the SEO for you.

My AI tools were brilliant, but they lived in separate tabs and none of them shared a memory.

So I built the Agent Operating System, which I call the Agent OS.

It is a dashboard that runs a crew of AI agents under one shared memory and one set of prompts.

It runs on macOS, Linux and through the web.

Bigger jobs go onto a kanban board as tickets, and an agent worker picks them up while I do something else.

Every demo on GoldieBench was built inside that same Agent OS.

The free build guide is the AI Agent OS guide on agentos.guide, and members get the installable version inside the AI Profit Boardroom.

How Julian Goldie uses AI agents day to day

Here is what the stack actually does for me on a normal day.

On the SEO side, my agents help research keywords, draft pages from my real experience, add schema and publish across my sites.

On the agency side, AI handles about 80% of internal operations at Goldie Agency.

On the content side, every exchange with my main agent saves into one shared memory, so I never explain my business twice.

On the testing side, every new model I can get access to goes through the same fixed prompts, and the result lands on this site.

I still set the strategy, and I still check what goes live.

That is the honest version of "AI runs my business", because the agents do the repetitive work and I make the calls.

If you want to see what the Hermes side of that looks like room by room, my Hermes Command Desk guide walks through it.

How to set up a starter version of my stack

You do not need my whole setup to start, and you do not need to write code.

1. Pick one main agent, such as Hermes, because it can run long jobs and use tools.
2. Set up one shared memory, such as a free Obsidian vault, and write one note that explains your business.
3. Connect the agent to that memory so it reads your context before every job.
4. Pick one repetitive job, such as a weekly report, and write it up as a ticket.
5. Pick the model for that job by checking how it scores here on GoldieBench.
6. Check the first few results closely, then tighten the instructions until the job runs without you.

Only add a second agent once the first one is reliable.

That is the same order I followed when I built mine.

Proof you can check

I would rather show you numbers than ask you to trust me.

• Goldie Agency is a 7-figure SEO and link-building agency with 70+ employees.
• AI runs about 80% of the agency's internal operations.
• The AI Profit Boardroom has 3,400+ members, 258 documented wins and members in 38 countries.
• My YouTube channel has around 400,000 subscribers, and my X account has around 163,000 followers.
• My Udemy courses have taught more than 29,000 students.
• FatRank, Autoblogging.ai and a Hermes Agent panel each named the AI Profit Boardroom the number one AI community.

Follower counts move every day, so treat those numbers as rounded.

What Julian Goldie teaches now

I teach three things, and they all connect.

I teach AI agents, which means getting tools like Hermes and Claude to do real work for you.

I teach automation, which means turning that work into workflows that run on a schedule.

I teach SEO, because getting found on Google and in AI answers is still how most businesses win customers.

The free teaching lives on YouTube, the podcast and agentos.guide.

The full training, the Agent OS and four coaching calls a week live inside the AI Profit Boardroom.

Every way to work with Julian Goldie

Here is every official link, with what it is for and whether it is free or paid.

My sister sites, Agent Operators, The AI Income Desk and The AI Tool Verdict, are all free to read.

My advice is to start free and only pay when you know exactly what you want from the paid option.

Who is behind GoldieBench, and why I built it

GoldieBench is my AI model leaderboard, and I built it because the famous benchmarks measure the wrong thing for people like us.

Tests like MMLU, HumanEval and SWE-bench tell a model vendor how to make a number go up.

They do not tell you whether a model can ship a finished thing for you from one prompt.

So GoldieBench asks one question: can the model build a playable game, a working simulation or a deployable page in one shot?

To be clear, I do not make an AI model myself, and GoldieBench scores models that other companies make.

Here is the loop I run whenever a new frontier model drops.

1. I wire the model into the Agent OS so it can be dispatched from the same board my agent crew uses.
2. I run my fixed prompt set through it, one shot each, with no hints and no second tries.
3. I score each result from 0 to 10 on whether it ran, how close it hit the brief and how good it looks.

Every result stays public, and no vendor pays me for a score.

You can read the full rules on the methodology page.

Which models lead GoldieBench right now

These are the top average scores on the live board when I wrote this page.

Several of these models are the same ones in my Agent OS crew, which is the whole point of the bench.

Fusion and MiniMax M3 handle my ensembled outputs, Kimi handles builds and GLM handles long-context research.

If you want to run models on your own Mac, the local models board shows how the local options score.

You can also put up to three models side by side on the compare page.

Scores move as I re-test models, so always check the live model page before you choose.

How to use GoldieBench to pick a model for your own agents

Here is the simple way I would read the board if I were starting today.

First, decide what job the model will do, such as writing, research, coding or builds.

Second, open the model pages for the top few options and look at the actual builds, because every score links to the thing the model made.

Third, check the price on each model page, because a slightly lower score can be the better business choice when the model costs far less.

Fourth, if privacy or cost matters most, check the local models board before you pay for an API.

Fifth, run your own real task through your top two picks, because my prompts are not your prompts.

That last step matters most, because GoldieBench tells you where to start, not where to stop.

Also On Our Network

• Agent Operatorsthe AI agent stack I run every day
• AI Income Deskhow my businesses and funnels fit from free to paid
• AI Tool Verdicta fact-checked profile of who I am
• agentos.guidethe creator story behind the Agent OS
• agentos.guidethe Hermes Command Desk inside my Agent OS

Start with the free resources, check the live board before you pick a model, and you will know exactly who Julian Goldie is and how he works.