When Mubbu recorded a few words about Andrew, he went straight to the scale of it.
"I’ve never seen an individual as smart as Andrew when it comes to AI. This guy has GPTs after GPTs created, and we’re working with him side by side to innovate, to be a company that no one else has seen."
Then he said something that’s easy to miss.
"We’re working AI, we’re working org structures, we’re working on our own operational systems."
AI was the first item on that list.
It wasn’t the last.
Hands on
Mubbu’s team made content for clients. Writers. Strategists. Real people learning each client’s voice and turning it into posts.
And they were proud of how hands on they were.
That pride was also the problem.
During a long working session with Andrew, the CEO in the room put it plainly:
"Me as the CEO, I look at that and I am like, There is a certain amount of efficiency that just isn’t there because of how hands on we are. Because of how manual every single thing is."
Every new client meant interviews and questionnaires. Every answer had to be read, sorted and turned into briefs. Every brief had to become content.
And nearly all of it was done by hand.
Maybe you know that tension. The thing that makes your business good is the same thing that makes it hard to grow.
The question
Andrew didn’t start with a grand redesign.
He started with one of the questions the team asked new clients.
It was a perfectly normal question: what are your main offers that you want to promote on Twitter?
The kind of question every agency asks.
And the kind that gets a flat answer.
So Andrew asked ChatGPT to rewrite the onboarding questions. The first attempt came back bloated.
"This is giving me way too much."
So he narrowed it.
"I am not really looking for you to completely redesign the entire process. What I want you to do is specifically take the questions that are being asked and I want you to ask them in a way that is going to actually resonate with our core demographic and get them speaking conversationally and passionately about the thing that they love versus just asking them basic boring questions like, what’s your unique selling proposition? What’s your product? What’s your target market?"
This is what came back:
"What’s the one offer or service you are most proud of and can’t wait to share with the world? Why does it excite you so much?"
Someone in the room reacted out loud.
"Wow, that’s a hard thing to pick."
That was the point.
Andrew explained why it worked:
"It’s invoking an emotion out of the person, right? What are you most proud of? And then that person is going to talk about what they’re proud of."
A flat question gets a flat answer.
A question about pride gets a story.
And stories are what content is made from.
Who does what
Here’s where the conversation stopped being about prompts.
Because a better question is useless if every strategist on the team has to learn how to write it.
Andrew was clear about what AI should and shouldn’t do.
"You can use ChatGPT for what it’s really good at, which is all of that busy work. And now use it collaboratively, because we’re not trying to replace your writers or your strategists. We want to make them more efficient."
And then he named what was really changing.
"I feel like with AI, it changes the role of the operator as well."
The team didn’t need to become prompt engineers. They needed a system someone had already built for them.
"It would take forever for me to teach every person on your team how to do what we’re doing right here. But we don’t need to because it only really needs to be done once. All they need to do is follow the instructions, not create it."
So the new flow looked like this.
Ask the better questions live, on a call. Take the call transcript. Feed it into a GPT built for that one job. Let it pull out the answers and put them where the team could use them.
The writers kept writing.
The busy work moved.
Andrew even had a name for the part that mattered most: keeping a person in the loop.
"The biggest mistake is trying to just automate your entire content creation process."
The voice
Getting the client’s voice right was its own puzzle.
Andrew showed them an example from a voice profile he’d built. He had told an AI the voice swears, but with intention.
"You don’t swear just to be a trucker. You swear because it drives home a point."
Then he asked ChatGPT how it would describe that instruction. It gave him its own wording:
"Profanity is used sparingly and intentionally only to underscore passion and emphasize a critical point."
Then he fed that back in.
That’s not how a person would say it.
It’s how the machine understands it.
And that small translation is a lot of what separates people who get generic output from people who get their own voice back.
The spare tire
Then they talked about the hardest part of the client relationship.
The month when clients go quiet.
They get busy. They stop answering questions. They stop reviewing drafts. And once they fall off, it’s hard to get them back.
The team could use AI drafts in each client’s voice, trained on that client’s own transcripts, to keep things moving.
But Andrew didn’t want that to become the business.
"We want to use this as our spare tire in the trunk, not as the tire we drive around on every day."
Use it when you need it.
When you don’t, go back to the real conversation.
He also wanted clients warned at the start that the work would feel amazing, then hard, then harder, before it got better. Knowing the pattern exists makes it easier to plan for it.
That’s not a prompt.
That’s how a company decides to treat its clients.
Org structures
That’s what Mubbu meant.
"We’re working AI, we’re working org structures, we’re working on our own operational systems."
The GPTs were real. There were plenty of them.
But the work was about who asks the questions, who reads the answers, where the transcripts go, what the writers stop doing by hand, and what happens when a client stops answering.
Tools sit inside a company.
Andrew was working on the company.
Mubbu could already see where it led.
"I can’t believe how efficient our work’s gonna be."
The stage
Months later, Andrew was standing in front of a room of founders in Miami.
He asked how many of them had an onboarding form for their agency.
Then how many asked a question like what’s your offer?
He told them what he’d found going through this work with Mubbu’s team. Ask someone to type out their offer and you get the bare minimum.
So he’d stopped using onboarding forms altogether.
"So, the way that I do those now instead is an actual live Zoom call. Even if you do that Zoom call in a group setting, it still ends up coming out better."
Better questions. Asked out loud. Captured on a recording. Pulled out with AI.
The same idea that had changed one question on Mubbu’s onboarding sheet was now part of how Andrew taught a room full of business owners to run their own.
Mubbu had said they were playing in a different league.
What they built together, Andrew kept using.