A data professional recently built a personal fitness tracking web app called Trainlytics in 4.5 hours using LLM agents and a spec-driven development approach. The project addresses the challenge of balancing running and strength training for half marathon preparation—a common pain point for athletes who find existing tools too fragmented.

The development method represents a significant shift from "vibe coding," a term Andrej Karpathy coined in February 2025. Karpathy himself acknowledged just one year later that the vibe coding era is ending, giving way to "agentic engineering"—where developers orchestrate AI agents against detailed specifications rather than writing code directly.

Spec-driven development addresses core issues that plague vibe coding: context decay across sessions, lost reasoning from AI conversations, and inconsistent team conventions. The approach involves creating a "constitution" with mission, tech stack, and roadmap documents before implementation. The developer used Visual Studio Code with the Claude Code plugin to build the app, following practices from JetBrains' DeepLearning.AI course on spec-driven development.

This approach decouples specification from implementation, preserving context across sessions and ensuring both humans and AI agents align around project non-negotiables—transforming AI-assisted development from experimental prototyping into structured engineering.