Building production ready full stack apps with AI

Key Concepts

• AI App Builder: A platform that enables users to create, test, and deploy production-ready software applications using AI agents.
• 0 to 1 Innovation: The process of creating something entirely new from scratch, rather than iterating on existing products.
• Coding Agents: AI systems capable of writing, debugging, and deploying code, often performing at a professional developer grade.
• Model Orchestration: The practice of dynamically routing specific tasks to the most suitable AI model (e.g., using Gemini Flash for speed/UI and Gemini Pro for complex logic).
• Experiment Velocity: The speed at which a company can test, measure, and iterate on product features or model performance to achieve statistically significant results.
• Manual vs. Automated Evals: The dual approach of using human feedback (vibe checks) and automated testing to validate AI model performance.

---

1. Main Topics and Key Points

• Democratizing Software Development: Emergent aims to bridge the gap for non-technical domain experts, allowing them to build full-stack applications without traditional coding skills.
• Production-Ready Output: Unlike many "no-code" tools, Emergent focuses on shipping apps that meet professional engineering standards, including backend, database, and UI integration.
• Infrastructure Strategy: The company built its own sandbox container technology on top of Google Kubernetes clusters rather than relying on third-party sandboxes, allowing for real-time, local feedback loops when agents encounter errors.
• Efficiency over Token Cost: A key insight presented is that more intelligent models often reduce total costs by completing tasks faster and with fewer errors, proving that "price per token" is a misleading metric for efficiency.

2. Real-World Applications

• Small Business Empowerment: A real estate professional in Norway used the platform to build an "Airbnb-style" rental property application over a weekend, replacing what would have been a $100,000 outsourced development project.
• "Wingman" Concept: An upcoming feature that allows users to build and iterate on applications via text-based prompts while on the go.

3. Methodologies and Frameworks

• The "Idea Guy" Evolution: The transition from "idea guys" being viewed as non-executors to becoming powerful architects who can use AI to manifest products instantly.
• Model Segregation: Emergent uses a sophisticated orchestration layer to delegate tasks:
  • Gemini Flash: Utilized for its speed and tool-calling capabilities, serving as the primary interface for daily user interaction.
  • Gemini Pro: Reserved for heavy-duty, complex logic tasks.
• Experimentation Pipeline: Borrowing from Google’s internal culture, the team uses rigorous A/B testing and statistical analysis to determine if a new model improves business metrics like conversion and retention.

4. Key Arguments and Perspectives

• The "Bet on Agency": Madhav argues that developers should avoid over-engineering their systems. Instead, they should build with the assumption that AI models will continue to improve, allowing the system to delegate more agency to the models over time.
• The Importance of Evals: Madhav emphasizes that manual human evaluation is a critical "first gatekeeper" for any AI startup before scaling to automated testing.

5. Notable Quotes

• "The idea guy used to be a bad term... but thanks to AI, now if you are an idea guy, you're actually much more powerful." — Madhav
• "Sometimes when you see a model becoming smarter, it may on the surface seem that the price has increased, but because the model has become intelligent, it will actually spend less time." — Madhav

6. Data and Statistics

• Growth: In less than one year since public launch, Emergent has reached $100 million in annualized revenue run rate.
• Scale: The platform has facilitated the shipping of approximately 10 million applications with a user base of 9 million.

7. Synthesis and Conclusion

The conversation highlights a shift in the software development landscape where the barrier to entry is no longer technical skill, but the ability to effectively orchestrate AI agents. By focusing on "production-ready" output and building proprietary infrastructure, Emergent has successfully enabled non-technical entrepreneurs to execute complex ideas. The core takeaway for builders is to prioritize robust evaluation frameworks (both manual and automated) and to trust in the increasing agency of AI models rather than building rigid, over-engineered systems.