From Vibe Coding to Spec-Driven Development

2026-05-12 · Source: Towards Data Science · Field: Technology & Digital — Software Development & Engineering, Artificial Intelligence & Machine Learning · Depth: Intermediate, long

Summary

The article introduces spec-driven development (SDD) as an evolution from "vibe coding" in the AI era, emphasizing structured planning and documentation for building analytical tools and web applications. Coined by Andrej Karpathy in February 2025, "agentic engineering" describes orchestrating AI agents with human oversight against detailed specifications. SDD addresses the limitations of iterative, unstructured "vibe coding," such as context decay and lack of shared conventions, by decoupling specification from implementation. The author demonstrates SDD by building "Trainlytics," a personal fitness tracking web app, using Visual Studio Code with the Claude Code plugin. The process involves defining a "constitution" (mission, tech stack, roadmap), iterating through feature development (plan, implement, validate), and regular replanning, taking approximately 4.5 hours to create a usable product.

Key takeaway

For AI Engineers and Data Analysts building complex applications, adopting spec-driven development is crucial. This approach mitigates issues like context loss and inconsistent codebases inherent in "vibe coding," especially in multi-developer environments. By focusing on detailed specifications and architectural decisions upfront, you can ensure higher quality software and more efficient development cycles, shifting your role towards strategic oversight rather than direct code generation.

Key insights

Spec-driven development with AI agents offers a structured approach to building software, enhancing collaboration and maintaining context.

Principles

• Decouple specification from implementation.
• Preserve context across AI agent sessions.
• Prioritize human oversight and architectural decisions.

Method

Define a project constitution (mission, tech stack, roadmap), then iterate through feature phases: plan, implement, and validate. Regularly replan to align with evolving goals and update specifications.

In practice

• Use markdown specifications stored in the repository.
• Make all changes through the agent to maintain consistency.
• Employ a second agent for reflection and plan critique.

Topics

• Spec-Driven Development
• Agentic Engineering
• Vibe Coding
• LLM Agents
• Software Development Workflow

Code references

• miptgirl/trainlytics
• github/spec-kit
• github/spec-kit

Best for: AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Extra #4 - Beyond “Vibe Coding”: Evolution of AI Development — Spec-Driven Development (SDD) offers a structured approach to manage LLMs, moving beyond unstructured "vibe coding."
• AI 101: From Vibe Coding to Spec-Driven Development — Spec-Driven Development uses a formal specification as the source of truth for AI-assisted code generation and validation.
• Spec-Driven Development: AI Assisted Coding Explained — Spec-driven development provides a structured, SDLC-aligned approach to AI-assisted coding, reducing ambiguity for LLMs.
• Spec-driven development: The AI engineering workflow at Notion | Ryan Nystrom — AI agents streamline engineering workflows, automating tasks from meeting prep to code generation and spec-driven development.
• DAY 5 Livestream - 5-Days of AI Agents: Intensive Vibe Coding Course With Google — Spec-driven development treats code as disposable, making specifications the durable, versioned source of truth for AI agents.
• The Spec Growth Engine: Spec-Anchored, Code-Coupled, Drift-Enforced Architecture for AI-Assisted Software Development — The Spec Growth Engine uses machine-enforced architecture to prevent AI context overload and spec-code drift in software development.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards Data Science.