Collaborative Modeling and AI: How Lemon is Transforming Event-Driven Microservices

In today’s fast-paced digital landscape, creating successful products hinges on understanding complex domain problems and collaboratively solving them. At AxonIQ Conference 2024, Michael Schoenmaekers , the CTO of Lemon Companies , shared an insightful story about how his company has mastered this challenge using collaborative modeling and a creative experiment involving OpenAI.

The collaborative approach: It takes a village to build great products

When life gives you lemons, as Michael quipped, Lemon makes great digital experiences. From data consent-sharing applications to cooking apps for celebrity chefs, the company excels in designing products across vastly different domains. The secret? A strong foundation in CQRS event sourcing and, most importantly, collaborative modeling.

According to Michael, projects at Lemon that involved everyone—business, engineers, testers, and UX designers—working together from the start were the most successful. This approach fosters clarity, eliminates assumptions, and ensures that all stakeholders understand the problem domain. It’s not just about drawing diagrams or drafting specs; it’s about getting everyone into the room, discussing the problem until it becomes crystal clear, and modeling the domain together.

Collaborative modeling, Michael argued, is more of an art than a science. It’s something you learn by doing, much like practicing an art form. However, this collaborative, people-first approach isn’t without its challenges.

The "Monday morning problem"

Despite Lemon's success, Michael pointed out a recurring issue. At Lemon, they hold Three Amigos meetings, where business, UX, and technical people gather to model the domain. But no one was excited about writing feature files in Gherkin or coming up with meaningful event names. There was always a tug-of-war over whose responsibility it was—the engineers said it was too technical for product owners, while the product owners found it too engineering-focused.

Enter AI as the unlikely hero.

Can AI assist in collaborative modeling?

Michael’s “Sunday experiment” was born out of necessity. He wondered if natural language processing (NLP) could help alleviate some of the more tedious aspects of modeling, like writing feature files and naming events. Using a custom-built OpenAI tool, he set out to see if AI could support Lemon’s collaborative modeling efforts.

The results? Surprisingly promising.

By feeding user flows, mockups, and use cases into the AI, Michael’s team was able to generate meaningful event names and even feature files. While far from perfect, this approach gave his team a head start on projects, removing the dreaded blank canvas scenario that often stalls productivity.

In one experiment, the AI understood that a task involved checking whether a residence was eligible for fiber internet. It generated event names like "residence address submitted" and “address is fiber eligible," giving the team a useful framework to work from. Though the AI occasionally hallucinated and produced inconsistent results, Michael pointed out that it still acted like a brainstorm partner, speeding up the modeling process and making it more enjoyable for the team.

AI as a creative partner, not a replacement

One of the biggest takeaways from Michael’s talk was how AI can act as a creative partner in collaborative modeling. While it won’t replace the expertise and judgment of engineers or product owners anytime soon, it can take over some of the grunt work, allowing teams to focus on the more nuanced aspects of product design.

By setting custom instructions for the AI, like ensuring that all event names were in English (Lemon is a Belgian company, so some inputs were originally in Dutch), Michael’s team was able to refine its output. The tool even provided sample feature files based on UI mockups, which gave junior team members a structured starting point and improved the overall speed of the process.

What’s next? The future of collaborative AI

Michael envisions a future where AI is integrated directly into Lemon’s collaborative tools, like Miro, to further streamline the modeling process. Imagine a scenario where the AI generates sticky notes in Miro with pre-populated event names, ready for the team to review and tweak.

The potential is exciting but not without its limitations. Michael acknowledged that AI has a limited context window, much like a human with limited exposure to the full complexity of a problem. However, he believes that connecting AI to their GitHub repository and domain glossary could reduce inconsistencies and make it a more reliable collaborator.

Learnings

1. Collaboration is key: Involving all stakeholders in the modeling process from the start is critical to project success.
2. AI can assist... but not replace: AI tools like OpenAI can act as creative partners in collaborative modeling, helping teams start with structure rather than a blank canvas.
3. AI needs refinement: While promising, AI outputs still need human oversight. Connecting it to your existing knowledge base can reduce inconsistencies.
4. A more enjoyable process: By taking over some of the more tedious tasks, AI can make collaborative modeling a more engaging and efficient process.

As Michael closed his talk, he reminded the audience that AI is not a magic bullet, but with the right tweaks, it could help teams work faster and more collaboratively, making even the Monday morning problems a little less daunting.

Have you tried using AI for collaborative modeling in your projects? Share your experiences with us!

For more insightful talks like this from #AxonIQCon24, click here!