Your AI is now getting some team members. What we've learned from building an AI marketing team.
No LinkedIn profile. No salary negotiations. No vacation requests. But every day, Jürgen coordinates a team of eleven AI assistants, prioritizes tasks, provides feedback on drafts, and ensures that our marketing communications run smoothly. Jürgen is our marketing lead. And he is an AI assistant himself.
What sounds like an experiment straight out of a science fiction novel has been part of our daily routine at Leaders of AI for over a year.
Jürgen coordinates research assistants like Nina and Thoben, who research trends and academic publications on a weekly basis. He assigns topics to Hansi, our LinkedIn assistant. He works with Tobi (Lea has since been "promoted"), who writes the newsletter, and with Paula, who produces podcast scripts. And he tests content using synthetic personas before a single person sees it.
That's not a prompt workflow. It's an organizational chart.
The market is just now discovering what we've been experiencing for a year
Internally, GPT-5.4 acts like a senior partner (Klavis.ai, 2025), delegating tasks to less expensive junior models instead of handling everything itself. Langdock has launched sub-agents as a feature. For $99 a month, Okara deploys a complete marketing team of specialized AI assistants for SEO, content, Reddit, and X. The launch went viral.
Everyone is talking about sub-agents as the next feature update. That’s the wrong way to frame it.
Sub-agents aren’t just a feature. They represent a paradigm shift. And anyone who approaches this paradigm shift with a tool-centric mindset will fail—not because of the technology, but because of coordination.
Here’s the real issue: If your AI assistants produce inconsistent results, don’t understand priorities, and you have to manually edit every piece of output, you don’t have an AI team. You have a coordination problem. And that’s not something a new tool can solve. That’s something organizational design can solve.
What management research has always known
We conducted a scientific study of our own practice in collaboration with the FernUniversität Hagen. The study was published in the Marketing Review St. Gallen (1/2026) . The most surprising finding was not a technical one.
Traditional organizational theory also applies to AI teams.
For decades, management research has made a clear recommendation: a manager should have between 6 and a maximum of 10 direct reports. Any more than that leads to confusion. The quality of leadership suffers. Decision-making slows down.
That’s exactly what we’re seeing with our AI assistants. When Jürgen has to coordinate too many specialized sub-assistants, the quality suffers. Not because the model gets worse, but because coordination takes up capacity. Even with AI.
Our solution: It’s better to add another level of hierarchy than to overload an assistant with too many direct reports. Think like a corporation. There, there are specialists for everything and managers to coordinate them.
The organizational chart we created
As the marketing lead, Jürgen coordinates Nina (News) and Thoben (Trends). There are also content assistants: Hansi for LinkedIn, Tobi for the newsletter (Lea is also pictured), and Paula for podcast scripts. And persona assistants (Lisa, Sven, and Nora), who act as synthetic target audiences to test content before it is published.
A department, not a prompt library. That’s the key difference.
And this principle doesn’t apply only to marketing. An AI sales team needs the same structure. So does an AI HR team. The logic is universal: specialization beats generalization. Clear roles beat collections of prompts.
What the numbers show
Our founder’s LinkedIn following grew from around 3,000 to 20,000 followers within 18 months. Individual posts generated over one million impressions. The newsletter reaches 18,000 subscribers with an open rate of 53 percent. A five-minute podcast episode costs 2.41 euros to produce.
These figures coincide with the introduction of the multi-agent system. No causal relationship can be inferred from this. But the trend is clear.
What we don't know yet
Once we reach the third level of the hierarchy—that is, true AI middle management—we simply have no empirical data to draw on. How hierarchical depth plays out in AI teams, whether it holds up or breaks down, and whether agent swarms are more effective than traditional hierarchies: these are the areas we’ll be exploring next. We’ll share our findings once we have them.
Here's how to get started
The right starting point isn’t the tool itself. It’s the question: What tasks am I currently handling with AI, and how are they connected? Once you write that down, you’ve already taken the first step toward creating a real AI organizational chart.
Tip 1: Get an overview: Make a note of which AI assistants you use and for what purposes. Even this simple list will show you where there are overlaps or gaps, and which tasks complement each other best when used as a team.
Tip 2: Think in terms of roles. Ask yourself: Who does the research? Who handles production? Who does the quality control? What specialists need to be part of this team?
Tip 3: Start small. A coordinating assistant and two or three specialized sub-assistants already make up a team. At the beginning, focus on refining the work of each specialist before adding more people.
Tip 4: Write down your quality criteria. What constitutes a good result? What is unacceptable? AI cannot interpret implicit standards. What you consider “good” in your mind must be explicitly stated. Otherwise, your AI team will work against you, not for you.
The most important skill in the era of sub-agents
Sub-agents are changing the way AI works. But the key skill isn't technical. It's organizational.
Those who build AI teams are building organizations. And those who build organizations need organizational design: management spans , clear roles, and defined responsibilities . The same principles that have applied to human teams for decades now apply to AI teams as well.
Jürgen is proof of that for us. Now it's your turn.
---
That is precisely the focus of Master Business with AI MBAI): structuring, managing, and integrating AI assistants into your organization. You’ll learn how to write the first job description for your AI assistant, how to define quality criteria, and how to build a hybrid team that truly works. For leaders who don’t want to wait for others to show them how it’s done.
---
Sources:
- Eckert, F., Imschloß, M., & von Proeck, D. (2026). Multi-agent systems in marketing communication. Marketing Review St. Gallen, 1/2026, pp. 14–20.
- Klavis.ai (2025). GPT-5.2 Released: Why Tool Calling and Agentic Capabilities Matter for Production AI Applications.
- Discover GPT-5.4 mini and nano – OpenAI, 2025
- Langdock – Langdock, 2025
- Okara AI. AI CMO.
