How I AI: Building a Real-World Business with Andrew Mason & Nabil Hyatt

Every time I think I have a handle on what AI is capable of, a story comes along that takes me somewhere completely new. This time, it wasn't a SaaS tool or a digital product—it was a brick-and-mortar retail business. I got to sit down with Andrew Mason, the founder of Groupon and now CEO of Descript, and Nabil Hyatt, a partner at Spark Capital. By day, they are a founder and a VC, but in their spare time, they've become something totally different: small business owners who used AI to open a real-life board game social club in Berkeley called Tabletop Library.

What I love about this story is that it completely flips the script on the idea that AI is just for software or huge tech companies. Andrew and Nabil showed me how they took a passion project—what I called a “slightly nerdy little bit niche hobby”—and turned it into a real-world business. As Nabil put it, “there's just no way this business would've existed without AI at about a hundred different levels.” They basically had a third co-founder, and that co-founder was AI. It handled everything from financial models and local permits to coming up with new ways for customers to have fun.

In our chat, they broke down three incredible workflows that got them there. First, they used Claude Projects as a central hub to handle strategy, planning, and research. Second, they created their own “Dewey Decimal System” for their massive collection of board games, a task that would have been a nightmare without AI. Finally, they built a custom AI concierge that connects players for game nights through a simple text message.

What's so cool about this is the mindset shift behind it. Andrew talked about “rewiring your brain” to remember that AI is there to help solve any problem, not just ones that involve code. This episode is such a great example of how AI can take the scary parts out of a passion project, handle all the tedious stuff, and bring ideas to life that might otherwise never get off the ground. I want to show you exactly how they did it.

Workflow 1: Building a Business from Scratch with an AI Co-Pilot

The hardest part of starting something new is that you don’t know what you don’t know, especially when you're jumping into a field where you have zero experience. For Andrew and Nabil, who are tech experts but not retail guys, AI acted as a business partner that could fill in all those knowledge gaps for them on the fly. Instead of spending months doing research, they could just ask direct questions and get full-blown strategies and documents back in minutes.

Step 1: Establishing a Central Brain with Claude Projects

Their whole operation lived inside a single Claude Project. You can think of it as the business's central brain or command center. Instead of having conversations scattered all over the place, every idea, question, and decision was fed into this one project. This set up a really effective feedback loop: they would ask Claude a question, it would create a document (like a business plan or a competitive analysis), and they would upload that document right back into the project to give it more context for the next conversation. Over time, the AI built up a deep, detailed understanding of their vision, goals, and limitations.

Step 2: Generating the Core Business Plan and Strategy

Once they had their AI co-pilot set up, they started with the basics. They didn't write some complicated prompt; they just started a conversation. Nabil said their first question was something as simple as:

I wanna open up a place to play board games with my friends in Berkeley. These are my goals.

From that simple start, they generated everything you’d need for a business plan:

• Financial Projections: Budgets, pricing models, and revenue forecasts to figure out if the business could actually work and not just be a “money pit.”
• Space Layout: Ideas for how to physically arrange the shop to fit the most tables while still feeling welcoming.
• Real Estate & Permitting: They gave Claude pages from Berkeley’s local ordinance websites to make sense of the complicated permitting process. It even helped them draft a letter of intent for their landlord.
• Marketing Materials: It also generated a draft PowerPoint deck they could show the landlord to prove they were a serious, well-planned business.

Step 3: Developing Customer Personas for Product-Market Fit

To make sure they were creating a place that would appeal to lots of different people, they used AI to build a framework for understanding their potential customers. They came up with a 3x3 grid that mapped players on two different scales: how dedicated they were to one game versus trying lots of new ones, and how introverted or extroverted they were.

This wasn't just a thought experiment. Claude helped them guess what percentage of each persona to expect and, even better, brainstorm a calendar of events designed for each group. This helped them create a business model that worked for everyone, from the hardcore Magic: The Gathering player to the social butterfly who wants to meet new people over a different game every week.

Workflow 2: Inventing a Dewey Decimal System for Board Games with AI

This next part is one of the most creative ways they used AI. They had to solve the “wall of games” problem. If you've ever been to a board game cafe, you know that feeling of being completely overwhelmed by all the choices. Andrew and Nabil wanted people to be able to discover games in their collection, just like you’d discover books in a library. Their solution was a custom Dewey Decimal-style classification system, a project that Nabil said would have been “literally impossible without AI.”

Step 1: Designing the Classification Framework

Working inside their Claude project, they designed the whole hierarchy for their system, which they named the Tabletop Library Classification System (TLCS). They came up with broad categories and gave them number ranges. For example:

• 400s: Cooperative Games
• 420s: Adventure Co-op Games

The system also included a decimal for how complex or “heavy” a game was, from .1 (lightest) to .5 (heaviest). So, a really complex adventure co-op game might get the code 420.5.

Step 2: Automating Categorization in Airtable

Once the system was designed, the real work began: classifying hundreds and hundreds of games. They put their whole library into a database in Airtable. Instead of manually researching each game one by one, they used AI to do it for them. By giving an AI model the game's name and the rules of their TLCS, it could automatically assign the right code to every game in their database. This probably saved them hundreds of hours of mind-numbing work and left them with a perfectly organized system.

Now, a customer can walk over to the 420 section and instantly see all the adventure co-op games, grouped together and sorted by difficulty. It creates a super intuitive way for people to find games that just wouldn't have been practical to build by hand.

Step 3: Curating Retail with AI-Powered Fields

They even used this categorization trick for the small retail section in their shop. Using Airtable's built-in AI features, they made curated collections, sort of like you’d see at an independent bookstore. They could define a category with a simple description like “Games about cats” or “Silent Strategists,” and AI would automatically fill that category with the right games from their inventory.

Workflow 3: Building a Text-Based AI Concierge for Player Matchmaking

The last big piece was figuring out how to make it easier for people to organize game nights, which can often be a huge pain. They imagined a service where a member could just send a text saying, “I want to play a deck-building game this weekend,” and an AI concierge would figure out the rest. They looked at a bunch of off-the-shelf tools, but nothing really fit what they wanted to do. So, they decided to build their own using a modern, low-code AI stack.

Step 1: Designing the Tech Stack

Their setup has three main parts that all work together:

1. Airtable: The database. This is where all their business data lives—users, tables, games, and reservations. The key was designing it with AI in mind from the start. They used flexible free-text fields for things like “gaming preferences” and “availability,” knowing that would be easy for a language model to understand.
2. n8n: The workflow automation backend. This is basically the agent's brain, connecting all the services and running the logic.
3. Twilio: The user interface. This gave them the phone number for members to text, which kept the experience for the user incredibly simple.

Step 2: Building the Agentic Workflow

Here’s how the concierge actually works:

1. A member sends a text to the Tabletop Library number. Andrew showed me this real example from their demo:

Hey, can you set up a game for me anytime this weekend? Maybe a deck building game.

1. Twilio gets the message and sends it to an n8n workflow.
2. The workflow first checks the user’s phone number against their Airtable members table to see who it is.
3. The user's message then goes to a central AI agent inside n8n. This agent can use a set of “tools,” which are really just functions that let it talk to the Airtable base (like create_record or get_records).
4. With a detailed system prompt as its guide, the agent reads the user's plain-English request, figures out which tool it needs, finds other members with similar interests, and gets a game group organized.

Step 3: Prompting the Agent with Plain English

But here's the really cool part: the instructions for this agent are written in plain English. Andrew showed me the system prompt and pointed out that it just tells the AI what to do, without any complicated code. As he put it, “You're not calling a function or anything like that. You're just literally writing down what you think it should go do, and then it'll figure out how to find the function and use it.” It's a great example of how these kinds of agentic workflows are making software development easier for more people to get into.

A New Era for Passion Projects

Andrew and Nabil’s story with Tabletop Library is so much more than a fun side hustle. It’s a road map for how anyone can use AI to build something in the real world—things that might have felt out of reach before. From the big-picture business strategy all the way down to their custom matchmaking service, every step shows how AI can act as a tireless co-founder, a sharp strategist, and an endlessly patient assistant.

This project really forced them to rewire their brains and change their default approach to solving problems. As Andrew said, the best way to really understand what AI can do is to start something from scratch where you couldn't do it otherwise. Their story is a great reminder for all of us to look at our own “crazy ideas” and ask ourselves: what if AI is the thing that could finally make this possible?

Brought to you by

• Lovable—Build apps by simply chatting with AI
• Persona—Trusted identity verification for any use case

Episode Links

• Watch on YouTube
• Listen on Spotify
• Listen on Apple Podcasts