.svg)
Why “AI as a colleague” is more than a catchy metaphor
Generative AI is no longer confined to tech teams: 78% of global organisations now use AI in at least one business function, and 71% rely on Gen AI specifically (Mckinsey). Yet, most employees still interact with AI transactionally (“Write me a post”). Mark van Horik argues the real magic appears when you treat AI like a coworker: sharing context, debating ideas, and refining outputs together. That shift turns efficiency gains into effectiveness gains: better ideas, sharper copy, richer insights.
AI is the extra part of my creative brain.”
Mark Van Horik
Strategic Marketing Consultant, Marketing Guys
Mark doesn’t just use AI; he collaborates with it. Rather than using a one all-purpose chatbot, Mark created a nine-persona AI squad that consists of nine specialized AI personas with names, responsibilities, and even unique personalities. From Ava (B2B content specialist) to Luna (visual designer), each AI has a defined role. Each is a project-based Claude AI assistant trained to perfection with role-specific prompts, past work examples, tone-of-voice guidelines, and clear ethical instructions that generate high-quality outputs. They are treated like real teammates, ingrained into workflows and introduced to clients.
AI is a colleague now. You must learn to collaborate with it, not fear it.”
Mark Van Horik
Strategic Marketing Consultant, Marketing Guys
The concept mirrors a fast-emerging trend called agentic AI: autonomous digital teammates that orchestrate workflows and make decisions independently. Many predict these agents will soon handle everything from marketing personalisation to virtual finance analysis, reshaping org charts along the way.
How to build your own AI “team”
Most people treat AI like a vending machine: input prompt, get answer. But Mark van Horik’s method flips that model on its head. Here’s how to do the same:
- Define specific roles: Identify areas within your operations where AI can add value, such as content creation, data analysis, or customer engagement. Then assign each assistant a clear scope of responsibility, just like you would with a human teammate.
- Choose the right platform: Mark built his AI team using Claude Projects, but you can also use OpenAI’s Assistants API, Google Gemini Gems, or any other platform that lets you define persistent assistants. Pick one that fits your tech stack and offers granular control.
- Provide comprehensive context: Generic prompts lead to generic results. Feed each assistant detailed background: tone of voice, job description, target audience, preferred frameworks, company guidelines, and trusted data sources. Mark’s copywriting assistant, Ava, was trained as a bilingual B2B SaaS writer fluent in CTA strategy and persuasion tactics; not just “an AI that writes blogs.”
- Establish trust boundaries: Tell your assistants when not to answer. Set guardrails for uncertainty. For instance, if Ava isn’t sure, she’s instructed to ask for clarification or flag the task; never hallucinate. That kind of boundary creates confidence and builds reliability over time.
- Implement feedback mechanisms and monitor performance: Review outputs regularly and fine-tune based on results. Like any good teammate, your AI assistants should improve with feedback. Store successful examples, iterate on what doesn’t work, and gradually raise the floor of performance.
- Be consistent: Stick with the same assistants for recurring work. This builds a sense of shared memory, reducing ramp-up time and letting each assistant “learn” your style, logic, and preferences. Less rebriefing, more results.
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
