• Wolfflow AI
  • Posts
  • 🐺 A Global Ad-Machine with Multi-Agents

🐺 A Global Ad-Machine with Multi-Agents

I built it with Python and LangChain. It might help you too.

Author

Paulo Dores
September 29, 2025

Hello,

This week, we are checking how to generate ads that target specific events around the globe, so that global companies like Coca-Cola can start reaching even more people with hyper-localization. The cool idea here is how to use that system to apply to your organization.

If you want to skip the whole text, let me TL;DR it for you:

Harajuku’s Iconic Kawaii Monster Café is Coming Back – as an Amusement Park

Image generated with GPT-1-Image

Event Summary

The Kawaii Monster Café, known for its vibrant and whimsical atmosphere, is reopening as an amusement park in Harajuku. This event is expected to draw large crowds, especially among young people and tourists, due to its unique and colorful attractions.

Marketing Potential

  • Score: 8

  • Reasoning: High potential. The event is visually striking and aligns well with Coca-Cola's values of happiness and togetherness. It targets a youthful audience and offers numerous opportunities for creative, colorful marketing campaigns. The immediate reopening adds urgency and timeliness.

Creative Brief

  • Marketing Angle: Leverage the vibrant and whimsical atmosphere of the Kawaii Monster Café amusement park to position Coca-Cola as a natural companion to this colorful experience, embodying the spirit of togetherness and joy that the park represents.

  • Target Emotion: The exuberance and wonder of stepping into a fantastical world full of color and imagination, where every moment is a delightful surprise shared with friends.

  • Key Message: Coca-Cola brings a splash of real magic to every colorful adventure at the Kawaii Monster Café amusement park.

  • Target Audience: Young adventurers and tourists exploring the vibrant attractions of Harajuku's Kawaii Monster Café amusement park, seeking unique and joyful experiences.

Text Assets

  • Slogan: カラフルな魔法を、一緒に。

  • Social Media Post: 🌈🎉 カワイイモンスターカフェでの冒険には、コカ・コーラが欠かせない！一緒に色とりどりの魔法を体験しよう。#カワイイ #コカコーラの魔法 #ハラジュクアドベンチャー

  • Web Banner Copy: ハラジュクのカワイイモンスターカフェで、友達と一緒に色とりどりの魔法を体験しよう！コカ・コーラは、あなたの冒険にリアルな魔法をプラスします。🎈✨

In the code, there are also two more sources to fetch the events from: São Paulo and New York. You can add as much as you want.

How can a global brand like Coca-Cola target local events around the world to advertise in?

I tried solving this issue by creating a system with multiple AI agents that research local events, compile information, assess which events make sense for that brand, then generate an image for an ad and the ad copy itself.

The result is an end-to-end creative pipeline that covers the entire creative process, from raw data ingestion to the creation of final text and image assets.

It is modular and extensible. Each agent is a self-contained module, making it easy to add new capabilities or agents to the pipeline.

These are the agents I created:

  • EventScanner: The "eyes" of the operation. Scrapes local news and event websites for potential opportunities.

  • TrendAnalyzer: The "strategist." Analyzes each event against Coca-Cola's brand values and assigns a marketing_potential_score.

  • CreativeDirector: The "brain." Takes high-potential events and creates a strategic brief defining the campaign's angle, emotion, and message.

  • Copywriter: The "voice." Generates compelling, localized ad copy, slogans, and social media posts based on the creative brief.

  • VisualStrategist: The "art director." Translates the creative brief into a rich, detailed prompt for a text-to-image model.

  • ImageCreator: The "artist." Executes the image prompt, generating the final visual asset for the campaign.

  • MarkdownWriter: Creates the publications in Markdown format, to simplify the task of posting them on Social Media.

Every Agent has their own system prompt. Then, we combine these system prompts with the request that comes from the other agents to pass it to the LLM (gpt-4o) so that it can complete its task.

As we build our workflow with agents, we can see how they are connected, how they can be improved, and which other agents we might add to make the system even more on autopilot. One agent that would be really cool to add right away is the Publisher Agent. It would use social media APIs to automatically post the approved content to platforms like X (Twitter) or Instagram.

Conclusion

The result is an ad machine that targets multiple locations and creates ads that fit specifically in that niche. We can improve the system to fit our particular use cases and help us impact more people.

❇️ If you liked the idea and want me to help your organization create a similar system, reply to this email. Or join us in the AI Agent Development Training to empower your whole developer team on how to create similar agents.