Descriptions:
Web Dev Cody walks through how his agentic coding workflow has transformed over the past few months, centering on a single headline shift: he no longer opens a text editor or uses tab completion at all. Instead, he relies entirely on natural language prompts to agent-based tools — primarily Cursor 3 and CodeEx — to read, modify, and create files across multiple projects from a centralized view. He argues that the ability to context-switch quickly between repositories has become more valuable than any autocomplete feature.
A standout section covers real-world model comparisons. When Opus 4.7 failed to fix a sticky layout bug after five attempts, and GPT 5.4 also couldn’t resolve it, Gemini 3.1 solved it on the first try. Cody uses this to argue that no single model is a silver bullet, and knowing when to switch is a core skill for modern developers. He also discusses asynchronous agent workflows for CI/CD failures, using GitHub Copilot for merge conflict resolution, and cloud-based agents (including Claude Code) that spin up virtual machines to fix failing tests and commit the result without human intervention.
The video closes with a deep focus on “skills” — custom instruction files embedded in the codebase that enforce standards like test coverage, separation of concerns, and security audits before a pull request is even opened for human review. AI-powered review tools covered include Code Rabbit, Gemini Code Reviewer, Bugbot from Cursor, and Claude Code, with Cody recommending at least one or two be layered into any serious agentic pipeline.
📺 Source: Web Dev Cody · Published April 17, 2026
🏷️ Format: Workflow Case Study
