The Infrastructure Behind a 10-Minute AI Game Studio

The Infrastructure Behind a 10-Minute AI Game Studio

Adrian doesn't do cities like a tourist. He's spent years traveling the world living out of a suitcase, skipping the guidebook stuff in favor of Couchsurfing, neighborhood restaurants, and the kind of local detail you only get by actually talking to people. That habit — of digging into a place until it feels lived-in — turned out to be the seed for something unexpected: he built an AI system that generates a fully playable, location-based murder mystery game for any city, in about ten minutes. We talk about what Adrian built to make them possible — an infrastructure stack (including ElevenLabs for voice) that handles story, character, animation, and map generation, all orchestrated to turn "a city" into "a game" almost as fast as you can name the place. He walks through how the pipeline works, then demos the whole thing live, start to finish, on air. IN THIS EPISODE * Why a long-time traveler ended up building games instead of a travel app * The infrastructure behind the 10-minute game: story, character, voice (ElevenLabs), animation, and map generation working together * A live, end-to-end demo of a game being built in real time * Adrian's ambitions to scale this to millions of players — and the brand he's building around it PLAY IT YOURSELF Adrian gave AgentGTM listeners a code to try it out — https://play.detective.games/?code=AGENTAI [https://play.detective.games/?code=AGENTAI]

Build Apps Faster with AI: Measure Twice, Cut Once

EPISODE SUMMARY Kyle James — early HubSpot employee and multi-time entrepreneur who's started and sold a company — has become a remarkably prolific builder. In this episode he shares his approach to rapid application development with AI: plan first and let two AI models collaborate (and argue) to produce complex, reliable outputs fast. The throughline is that the hard part of building with AI isn't the coding — it's the thinking you do before you code. Kyle then puts the method to work in a live demo, building a competitive intelligence tool from scratch on screen. ---------------------------------------- GUEST Kyle James — early HubSpot employee, founder who sold a company to private equity, now builds AI agents and tools for clients and teaches his methodology rather than starting another company. ---------------------------------------- WHAT KYLE HAS BUILT * FromThePulpit — 30 years of his father's sermons turned into a searchable database you can chat with → https://www.fromthepulp.it/ [https://www.fromthepulp.it/] * CancerLogix — a comprehensive public directory of cancer research centers, publications, grants, and clinical trials → https://www.cancerlogx.com/ [https://www.cancerlogx.com/] * LootSignal — market intelligence for the retro game market; pricing, scarcity, desirability, and audience attention combined into a single Opportunity Score per title → https://loot-signal.vercel.app/ [https://loot-signal.vercel.app/] ---------------------------------------- THE METHOD: "MEASURE TWICE, CUT ONCE" Kyle's core argument is that a bulletproof plan makes for a smooth build — so most of the effort goes in before any code is written. Two ideas anchor the approach: * Plan before you build. Break a project into phases and milestones, and pressure-test the technical plan before execution. A solid plan is what makes the build go fast and stay (mostly) bug-free. * Two models, better together. Kyle uses Claude Code and Codex in tandem and leans on their different strengths — having them review and challenge each other's work, going back and forth until the plan and the code hold up. He thinks of AI models like very smart interns: powerful, but they don't always think everything through, which is exactly why the planning and cross-review matter. He's published his framework as a public, downloadable resource (see Resources below). ---------------------------------------- CHAPTERS * 00:00 — Introduction to Kyle's journey * 01:52 — From analog to digital: the FromThePulpit project * 06:25 — Planning for success: the framework behind fast building * 11:05 — Codex + Claude Code: better together * 15:37 — Live demo: building a competitive intel tool * 21:53 — Creating a structured plan * 25:43 — Drafting and reviewing the execution plan * 28:07 — Iterative development and adding features * 34:36 — Code review and quality assurance * 36:37 — Final thoughts and future applications ---------------------------------------- RESOURCES & LINKS * Trinity Skills (Kyle's planning + review framework, public repo) → https://github.com/kaileconsulting/trinity-skills [https://github.com/kaileconsulting/trinity-skills] * From the Pulpit code - Code to build a basic RAGG on top of a library of pdfs - https://github.com/kaileconsulting/fromThePulpit [https://github.com/kaileconsulting/fromThePulpit]

Building for Agents Before Agents Asked: The Koko Finance Story

Discover how Madan Chaukulkar [https://www.linkedin.com/in/madan-chaukulkar-184668/] has leveraged no-code tools and the MCP protocol to build an AI application for consumers and startups. This episode explores the evolution of Koko Finance from a web app to data available via API and MCP. We talk about: * The origins and purpose of Koko Finance in personal finance management * The concept of Model Context Protocol (MCP) and who would consider using it * Building a scalable MCP server for finance data and third-party integrations * Real-world use cases: rewards optimization, travel planning, and credit management * The roadmap for expanding Kokos reach into banks, startups, and credit unions Timestamps: 00:00 - Introduction to the Agent GTM podcast and Madan's background 00:55 - Overview of Koko Finance and its web app for credit card optimization 02:00 - Turning a personal project into a scalable MCP server for developers and startups 03:15 - The role of no-code tools like Claude Code in building and deploying applications 05:45 - Building an MCP server: technical approach and best practices 07:20 - Real-world use cases: rewards management, travel planning, credit monitoring 09:30 - Connecting MCP servers to Claude or GPT 11:00 - Managing and updating financial data with automation 12:30 - Strategic roadmap: targeting startups, regional banks, and issuers 14:15 - Madan's advice for developers interested in MCP and AI data integration 16:00 - Future of AI-powered financial tools and the importance of structured data Resources & Links: * Koko Finance Official Website [https://kokofinance.net/] * Madan's LinkedIn [https://www.linkedin.com/in/madan-chaukulkar-184668/] * Early Demo of Koko [https://youtu.be/Byp3Grm5QG4?si=SCulfWEeA3bhB6sE]

How Much Does Your Dream Life Actually Cost? Building an AI Calculator with Nico Krachenfels

What if you could calculate how much money you actually need to live your ideal lifestyle? In this episode, we sit down with Nico, an analytics manager, who created the Enough calculator — a personalized tool that figures out exactly how much money you need to live your ideal life. You'll discover why Nico decided to build this outside of his 9-5, what tools he used (and how accessible they are), how AI changed what a "cost of living calculator" could actually do, and what he's learned along the way. ---------------------------------------- KEY TAKEAWAYS * Building with AI is more accessible than you think. Nico used no-code tools like Lovable to create something real without being a full-stack engineer. If you can describe a problem, you can start building the solution. * People with full-time jobs are building things that matter. You don't need to quit to create. * Great side projects start with a real problem. Nico's calculator solves something he personally cared about. ---------------------------------------- TRY THE CALCULATOR You can try the Enough Calculator here [https://enoughmoney.life/]. Connect with Nico here. [https://www.linkedin.com/in/nicokrachenfels/]