Скачать или смотреть Spec Driven Development: NYT Connection Puzzle Assistant | Phase 4

In this video, we continue the Specification-Driven Development (SDD) journey using GitHub’s Spec-Kit and VSCode Agent Mode with GitHub Copilot, applying it to a real, end-to-end business problem: solving the New York Times Connections puzzle.
Building on earlier phases, this walkthrough focuses on Phase 4, where a significant new feature was delivered in under four hours using an agentic, human-in-the-loop workflow. The result is a major usability improvement: initializing a Connections puzzle directly from an image copy/paste, powered by an LLM-backed backend service.
This video shows how high-level specifications were drafted, reviewed against the project constitution, refined into detailed requirements, planned, implemented, and fixed — all using Spec-Kit commands and AI-assisted tooling, with deliberate human oversight at each step.
If you’re interested in AI-assisted software development, agentic workflows, or Specification-Driven Development, this is a concrete, real-world example that goes beyond demos and into production-style delivery.
Links to earlier videos in this series on GitHub Spec-Kit, how it works under the hood, and why it changes AI coding workflows can be found in the description.

📂 Project GitHub Repository
👉 https://github.com/jimthompson5802/sd...

Spec-kit videos:
   • The ONLY guide you'll need for GitHub Spec...  
   • GitHub Spec Kit DOES WHAT Under The Hood?  
   • Spec Kit: Github's NEW tool That FINALLY F...  

⏱️ Video Chapters
00:00 – Introduction
01:42 – Recap: NYT Connections puzzle & project overview
02:44– What changed in Phase 4 (problem & goals)
03:25 – Summary of process step durations
04:27 – Overview of high-level specification
05:26 – Estimated time comparison Manual vs Agent Development
06:40 – Pre-work before using spec-kit
11:12 – Detailed Specification Definition: /task and /analyze
12:28 – Create Technical Plan: /plan
16:14 – Detailed Task Definition: /tasks and /analyze
17:36 – Implement the change: /implement
19:58 – Code Fix up
21:47 – Demo of the Phase 4 Feature
30:28 -- Phase 4: Bottom Line
31:00 -- Next Enhancement

🚀 What Was Delivered in Phase 4
Frontend: Updated web UI to initialize puzzles from a pasted image
Backend: New LLM-powered service to extract puzzle words from images
Testing: Updated frontend and backend unit & integration tests (≥80% coverage)
Total time: ~3.75 hours end-to-end
Implementation time: ~56 minutes using Spec-Kit /implement
Human-in-the-loop: Targeted fixes resolved in ~12 minutes