After 30 Years in Management, He Built a Production App With AI
Fran Muñoz turned an AI coding experiment into “minimal”—a production application that people at Zyte now rely on for real delivery work. The application runs on roughly $70 per month in Google Cloud infrastructure. The platform it replaced had cost around $5,000 per month. But this story is about more than reducing a software bill. Fran had spent nearly 30 years in management rather than day-to-day development. By combining domain knowledge, systems thinking, and an AI coding agent, he built an application that now supports project portfolios, revenue forecasting and recognition, renewals, churn, infrastructure costs, dashboards, and automated workflows. In this conversation with Neha Setia Nagpal, Fran explains: • How a small experiment became production software • Why he let the AI agent work directly with the code • How specification-driven development preserved architectural decisions • Why approximately 3,600 automated tests gave him confidence to deploy continuously • How he turned repeated development processes into agent skills • Why technical instinct still matters when AI writes the code • How GitHub Copilot, Claude Sonnet, and Claude Code featured in his journey • Why human mentorship remains essential The $70 figure represents the application’s infrastructure cost. Internal ownership still carries responsibilities such as maintenance, testing, monitoring, and Fran’s time. CHAPTERS 0:00 Meet Fran and “minimal” 2:26 AI as a force multiplier 5:40 The domain knowledge behind the application 8:48 Why the previous platform wasn’t enough 13:34 How the MVP grew into a production system 19:40 From GitHub Copilot to Claude Code 22:44 Letting the agent touch the code 25:42 Specification-driven development 27:47 Approximately 3,600 tests and continuous deployment 30:37 Maintenance, refactoring, and agent skills 36:08 Teaching technical instinct 43:01 Domain knowledge, systems thinking, and AI leverage 47:13 Why a mentor can be better than a book 51:58 Running the application for roughly $70 a month 56:12 Tools, models, and estimated build time 57:45 How Fran creates and improves agent skills 1:02:44 The lesson Fran took from the journey RESOURCES FRAN LEARNED FROM Specification-driven development: https://youtu.be/3le-v1Pme44?si=Y2bgJkNVUBESV5JF Agent skills: https://youtu.be/4mnP1lRdUm8?si=bEwkqjDFUi-VRDsi RELATED ZYTE STORIES Treat your AI skills like software, starting with evals: https://www.zyte.com/blog/treat-your-ai-skills-like-software-starting-with-evals/ The best agent skill is the one that says the least: https://www.zyte.com/blog/the-best-agent-skill-is-the-one-that-says-the-least/ “I have superpowers now”: https://www.zyte.com/blog/not-an-interview-i-have-superpowers-now/ Building a web-scraping copilot: https://www.zyte.com/blog/adrian-chaves-web-scraping-copilot-interview/ Read the companion article: [https://www.zyte.com/blog/a-70-a-month-production-app-replaced-a-5-000-a-month-platform-what-production-grade-ai-coding-looks-like/] EXPLORE ZYTE Zyte on GitHub: https://github.com/zytedata Try Zyte API with free credit: https://www.zyte.com/sign-up/ Subscribe for more conversations about AI-assisted development, agent workflows, software engineering, and building dependable production systems. #AICoding #SoftwareEngineering #AgenticAI