How a VC and tech founder used AI to launch a brick-and-mortar business in their spare time

Youtube URL: https://youtu.be/HuLL6wOEIB8

Host(s): Claire Vo

Guest(s): Andrew Mason (CEO, Descript), Nabeel (Nabeel) Hyatt (General Partner, Spark Capital)

Podcast Overview and Key Segments

Overall Summary

This episode of How I AI features Andrew Mason and Nabeel Hyatt on how they used AI to launch Tabletop Library, a membership-based board game social club and retail space in Berkeley. They explain how AI made the venture viable despite their full-time roles. The team used Claude as: (1) a business partner to draft plans, research, and modeling; (2) a manual worker for grunt tasks like cataloging and labeling; and (3) an experience engine to power an AI concierge that forms game groups via SMS. Their stack includes Airtable as a human-friendly database, n8n for workflows, Twilio for texting, and Claude with MCP tools to act on structured data. They also built a Dewey-like catalog for hundreds of games and used integer programming for capacity planning. A core lesson: rewire your habits to “remember AI exists” at every step.

Reference

• Dewey Decimal System: A library classification method that groups items by subject.
• TLCS code: A custom “Tabletop Library Classification System” used to group games (e.g., 420.5 = Cooperative > Adventure > heavier complexity).
• LFG: “Looking for gamers” (like “looking for group”).
• MCP (Model Context Protocol): Lets AI tools call functions and operate on external systems like databases.
• Airtable: Spreadsheet-like database with views, filtering, and AI fields.
• n8n: Workflow automation tool that connects apps and APIs without heavy code.
• Twilio: Messaging API platform used to send and receive SMS.
• ILP (Integer Linear Programming): A method for optimizing discrete choices.
• PuLP and CBC: Python library (PuLP) and solver (CBC) used to run ILP models.
• Windsurf and Cursor: AI coding IDEs.

Key Topics

Why AI made a physical, local business viable

Andrew and Nabeel describe Tabletop Library as a project that “would not exist without AI.” They lacked small business and retail experience, and had no spare time for startup grunt work. AI removed those constraints. It generated business plans, financial models, and permitting checklists. It drafted landlord LOIs and decks. It guided space layout tradeoffs (even if spatial planning needed human judgment). Most importantly, AI sped up feedback loops so they could stay in creative mode while still moving forward. The result is a community space that blends membership, retail, events, and on-demand group formation. This is not AI replacing people. It is AI enabling a category of services that would be too slow or costly to offer by hand.

AI as business partner, manual worker, and experience engine

They used Claude in three roles. As a business partner, Claude produced plans, competitive scans, pricing ideas, personas, and timelines. As a manual worker, it categorized hundreds of games, generated labels, and populated Airtable at scale. As an experience engine, it powers an SMS concierge that forms game groups in the background. The team would brainstorm, then iterate with Claude through documents, not just chat. New artifacts were added into Claude Projects as living context. Over time, the AI had full project memory and could propose sharper answers, content, and decisions. This pattern—documents as evolving context—kept them moving fast.

Personas and programming events

They built a 3×3 persona matrix. X-axis: preference breadth (one-game loyalist vs. loves new games). Y-axis: social mode (introvert vs. extrovert). Claude estimated distributions across these personas and suggested event types for each. This led to programming that serves chess-only regulars, Magic: The Gathering fans, and social explorers seeking new titles and new people. The persona work fed revenue models and pricing, and also informed the SMS concierge logic for matching. The key insight: traditional game shops optimize for narrow slices (e.g., a single TCG night). Tabletop Library programs for the broader, mainstream rise of board gaming as a social activity.

Creating a “Dewey for board games” and curated merchandising

They invented a Dewey-like classification system (TLCS) to cluster games on shelves by type and weight. Example: 400 = Cooperative, 420 = Cooperative Adventure, .5 = heavier complexity. A shelf label like 420.5 helps visitors find games they will enjoy fast. They also curated retail sections with playful themes (“Brain Burners,” “Push Your Luck,” “Silent Strategists”). Airtable’s AI fields and Claude’s prompts made categorization and curation feasible at scale and easy to maintain. Without AI, this level of detail would be cut for time. With AI, it becomes a signature experience that eases discovery and reduces decision fatigue in a wall-of-games setting.

Building the AI concierge (SMS LFG) with Airtable, n8n, Twilio, and Claude

Customers text a number to request a game session. The agent checks Airtable for the member, parses the request, and recruits other members with matching interests and availability. If a quorum forms, it books a table and confirms. Stack: Twilio handles SMS. n8n orchestrates flows. Airtable holds members, preferences, availability, games, tables, and reservations. Claude, via MCP tools, reads and writes records and follows a large prompt that acts like routing instructions. The interface is simple (SMS), while the logic and data live in Airtable. This cuts staff workload and unlocks an experience that would be too slow or costly for humans to run by hand.

Real estate, permitting, and confidence-building

They still used brokers and lawyers. AI acted as a second opinion to reduce fear and speed decisions. They fed Berkeley ordinance pages into Claude, asked it to explain tradeoffs, and drafted LOIs and decks tuned to landlords. This release valve turned “Byzantine and scary” into tractable steps. AI did not replace experts. It made them more effective and kept the founders from stalling. The combo of expert counsel + AI research + targeted prompts gave them the confidence to sign a lease and navigate local process. Crucially, it helped them stay moving while avoiding obvious mistakes.

Capacity planning with optimization

To validate viability and pricing, they built a Python model that uses PuLP and CBC for integer linear programming. Inputs: personas, event programming, pricing plans, and utilization assumptions. Outputs: member caps, break-even scenarios, and revenue envelopes. This connected strategy (who we serve) to operations (how many tables, when, and at what price) and finance (can we break even?). AI supported the data production and documentation that feed the optimization. The result is not a guess. It is a testable plan they can tune as real data arrives.

Tooling and the “documents as context” workflow

They used Claude Projects as the central hub. Each plan, deck, persona set, pricing sheet, and prompt became reusable context. MCP tools connected Claude to Airtable so the agent could act, not just suggest. They used Cursor and Windsurf for coding, and Airtable for a human-friendly database. n8n tied the flows together. Twilio handled messaging. They also used built-in Airtable AI fields for categorization. The key practice is simple: put everything in the AI project, iterate in documents, and keep promoting artifacts into the project’s context so the AI keeps getting smarter about your business.

Key Themes

AI as a co-founder for small, local businesses

AI helped them model a retail concept, shape a business model, and handle the unglamorous work. It also enabled a novel, automated experience (the SMS concierge). This is a template for AI-first brick-and-mortar: use AI to plan, to execute, and to power unique services. It is not about displacing staff. It is about doing what humans cannot do at speed or scale.

• Quotes:
• • “There’s just no way this business would have existed without AI at about a hundred different levels.”
  • “It’s not about replacing a human job… It’s enabling experiences that just couldn’t otherwise exist.”

Rewiring habits: “Remember AI exists”

Both guests stress the mindset shift. Success came from treating Claude as the default starting point for research, planning, modeling, and creative iteration. They built a habit: “Have you asked Claude yet?” Documents became living context. This habit accelerated learning and drove better outcomes with fewer delays.

• Quotes:
• • “There’s something that needs to happen where you remember that AI exists as you are contemplating a problem.”
  • “Treat it as a muse instead of an oracle.”

Data-backed programming meets community needs

Personas, event design, and pricing connect to capacity planning. The catalog system reduces choice overload. The concierge reduces coordination friction. All serve the emerging mainstream of board gaming, not only hardcore niches. This is service design with data and AI at the core.

• Quotes:
• • “We ended up with this 3×3 matrix of personas.”
  • “We’re trying to cover for the full penality of people now playing board games.”

Simple interfaces, powerful backends

Texting is easy and universal. Behind it sits Airtable, n8n, Twilio, and Claude with MCP. The lesson: start with the simplest interface customers will use. Put complexity in the backend where AI can orchestrate it. This lowers staff effort while raising service quality.

• Quotes:
• • “All we’re talking about is a simple set of tables… and a chatbot on top.”
  • “You’re just literally writing down what you think it should do and it finds the function.”

Expert plus AI beats either alone

They still used brokers and lawyers. AI was the second opinion and research assistant. This combination gave them speed and confidence, especially on permitting, LOIs, and landlord decks. It kept momentum without cutting corners.

• Quotes:
• • “We still relied on experts all the way through.”
  • “AI was giving us the confidence… so we weren’t going to get fleeced.”

Key Actionable Advise

• Key Problem: You have a side-venture idea but no time or retail experience.
• • Solution: Use an AI project hub as your co-founder.
  • How to Implement: Create a Claude Project. Draft vision, pricing, personas, and timelines. Keep all artifacts in the project. Iterate as documents. Promote each artifact into project context.
  • Risks to be aware of: Hallucinations and generic advice. Cross-check critical items with experts.
• Key Problem: Coordinating customers is slow and manual.
• • Solution: Build an SMS concierge with an AI agent.
  • How to Implement: Use Twilio for SMS, Airtable for members/games/tables, n8n for flows, and Claude with MCP tools to read/write Airtable. Write a routing prompt that covers common cases and fallbacks.
  • Risks to be aware of: Message fatigue, privacy, and edge-case errors. Add opt-outs, logging, and manual override.
• Key Problem: Customers feel overwhelmed by choices.
• • Solution: Create an AI-powered classification and curation system.
  • How to Implement: Define a simple taxonomy (e.g., genre > sub-genre > weight). Use Claude/Airtable AI to assign codes and produce labels. Color-code by complexity. Curate retail themes.
  • Risks to be aware of: Inconsistent tagging. Build QA checks and allow staff edits.
• Key Problem: Real estate and permitting feel opaque and slow.
• • Solution: Use AI as a research aide and second opinion, alongside experts.
  • How to Implement: Pull local ordinances and landlord requirements into Claude. Ask for summaries, risks, and draft LOIs/decks. Validate with your broker and lawyer.
  • Risks to be aware of: AI misses nuance. Always confirm with qualified professionals.
• Key Problem: Unsure if the business can break even.
• • Solution: Run a basic optimization-backed capacity plan.
  • How to Implement: Define personas, event mix, pricing, and utilization assumptions. Use Python with PuLP/CBC to model membership caps and revenue. Stress test scenarios.
  • Risks to be aware of: Bad inputs lead to bad outputs. Revisit assumptions with real data.
• Key Problem: Tooling friction blocks non-technical staff.
• • Solution: Choose a human-friendly database as your source of truth.
  • How to Implement: Use Airtable for records, views, and filters. Add AI fields where helpful. Connect via n8n and MCP for automation and agent actions.
  • Risks to be aware of: Vendor limits, API quotas, and costs. Plan for scale or migration paths.
• Key Problem: Teams forget to use AI consistently.
• • Solution: Build the “Have you asked the AI?” habit.
  • How to Implement: Route all research and docs through your AI project. Use checklists that start with “Ask the AI.” Share prompts and artifacts in one place.
  • Risks to be aware of: Over-reliance on first drafts. Keep human review and local context in the loop.

Noteworthy Observations and Unique Perspective

• Treat AI as a muse, not an oracle. Use it to generate options, not dictate truth.
• • Quote: “Treat it as a muse instead of an oracle.”
• AI compresses time-to-iteration and keeps you in flow.
• • Quote: “It feels like you’re mainlining that… your feedback cycle is so fast.”
• Some features exist only because AI exists.
• • Quote: “Very unlikely any of these products would even exist… the whole project wouldn’t have existed without AI in the loop.”
• Simple UIs win when the backend is smart.
• • Quote: “It’s just a simple set of tables… and a chatbot on top.”

Companies, Tool and Entities Mentioned

• Companies/Orgs: Descript, Spark Capital, Tabletop Library, Lovable (sponsor), Persona (sponsor), City of Berkeley
• Tools/Tech: Claude (Projects, MCP), Airtable, n8n, Twilio, Cursor, Windsurf, Notion (mentioned), PuLP, CBC
• Games: Catan, Wingspan, Slay the Spire, Magic: The Gathering, Pokémon, Chess, Dungeon Mayhem, Sky Team

Linkedin Ideas

• Title: AI’s Three Roles in a Real-World Startup: Partner, Worker, Experience Engine
• • Main Point: Show how AI can plan, execute grunt work, and power new services.
  • Core Argument: This mix made a brick-and-mortar venture feasible for time-poor founders.
  • Key Quotes: “There’s just no way this business would have existed without AI…”; “We used Claude as a co-pilot, a manual worker, and to create new kinds of experiences.”
• Title: How an SMS Agent Filled Our Tables: The Stack Behind AI LFG
• • Main Point: Explain Twilio + n8n + Airtable + Claude MCP for on-demand group formation.
  • Core Argument: Simple UX plus smart backend boosts utilization with minimal staff load.
  • Key Quotes: “You’re just literally writing down what you think it should do and it finds the function.”
• Title: From Overwhelm to Delight: Building a “Dewey” for Board Games with AI
• • Main Point: Show how a classification system reduces choice overload and speeds discovery.
  • Core Argument: AI makes high-touch curation feasible for small teams.
  • Key Quotes: “Impossible, literally impossible without AI.”; “420.5… Adventure co-op games… heavier complexity.”
• Title: Rewiring Founder Habits: The Discipline of “Have You Asked the AI?”
• • Main Point: Habit, not hacks, unlocks AI’s compounding value.
  • Core Argument: Make AI the first stop for research, drafts, and decisions.
  • Key Quotes: “Remember that AI exists…”; “Treat it as a muse instead of an oracle.”
• Title: Confidence in the Unknown: Using AI to Navigate Real Estate and Permitting
• • Main Point: AI as second opinion speeds learning and reduces fear.
  • Core Argument: Pair AI research with expert counsel to keep momentum.
  • Key Quotes: “We still relied on experts…”; “AI was giving us the confidence…”

Blog Ideas

• Title: The AI-First Playbook for Brick-and-Mortar: From Idea to Opening Day
• • Main Point: A step-by-step guide across planning, tooling, operations, and launch.
  • Core Argument: Small teams can ship ambitious physical experiences by defaulting to AI.
  • Key Quotes: “This business would not have existed without AI…”; “Documents as context” workflow.
• Title: Designing AI-Powered Community Experiences: Personas, Programming, and an SMS Concierge
• • Main Point: How to tie personas to programming and automate group formation with AI.
  • Core Argument: Service design + AI unlocks new forms of community at scale.
  • Key Quotes: “We ended up with this 3×3 matrix…”; “Go out and find this group for me…”
• Title: A Library for Games: Building a Classification System with AI That Customers Love
• • Main Point: Reduce decision fatigue with an AI-built catalog and curated shelves.
  • Core Argument: Practical taxonomy plus AI tooling turns a wall of games into a guided journey.
  • Key Quotes: “TLCS code 420.5…”; “Impossible… without AI.”
• Title: From Fear to Action: Using AI to Tackle Leasing, Permits, and Landlord Pitching
• • Main Point: Practical tactics for drafting LOIs, decks, and decoding local rules.
  • Core Argument: AI speeds learning and boosts confidence, but experts still matter.
  • Key Quotes: “We still relied on experts…”; “AI was giving us the confidence…”
• Title: Optimization in the Real World: Capacity Planning for a Membership Club
• • Main Point: How ILP with PuLP/CBC helps test break-even and member caps.
  • Core Argument: Tie personas and pricing to operations for data-backed decisions.
  • Key Quotes: “We built… an integer linear programming application to figure out capacity planning.”