Summary of "I Can't Believe This Actually Works"

Transformative Experience with AI-Powered Coding Agents

In this energetic and candid video, the creator shares a surprising and transformative experience with AI-powered coding agents—specifically “Ralph loops”—that completely changed their perspective on AI’s usefulness in software development. Initially skeptical and resistant to relying on AI for serious work, they recount how a push from a friend led them to build a small productivity extension. This project then snowballed into a multi-day deep dive where they spent nearly $300 on inference calls experimenting with AI-driven automation.

Highlights

Rapid AI-driven development: Over just a few days, the creator used AI agents to build and overhaul multiple projects, including:
- A Raycast extension
- A payment system
- Bug fixes
- A dashboard to monitor AI usage
- A client-server app with Elixir Phoenix They even prototyped a quirky “lift tracker” with minimal effort.
Ralph loops explained: The video dives into the concept of Ralph loops—simple while loops that repeatedly call AI models to iteratively build or improve code. This approach helps avoid “context rot,” which occurs when AI models degrade in quality due to bloated context from previous iterations. The creator compares their own implementation with a well-known engineer Joffrey Huntley’s approach, recommending viewers check out Joffrey’s hilarious yet insightful video.
Changing role of the developer: Rather than writing code line-by-line, the creator sees their role shifting to that of an orchestrator or “senior engineer” who directs AI “junior” agents. They emphasize the importance of:
- Clear planning
- Feedback loops
- Using statically typed languages to reduce AI hallucinations
Tooling and infrastructure: The creator praises platforms like Railway for fast, cheap app deployment. They also discuss various AI model subscriptions and APIs they’ve used, including OpenAI’s GPT-5.2 codecs and OpenCode Black, noting the high costs but incredible potential.
Reflections and future outlook: Initially grieving the end of traditional coding, the creator now feels empowered to build more ambitious projects faster. They urge viewers to push the limits of what AI can do, acknowledging that the full potential is still largely unexplored.
Jokes and personality: The video has a casual, conversational tone with moments of self-deprecating humor (“I was just skitso talking into whisperflow”), playful frustration (“please stop writing Ruby and [expletive] like that”), and awe (“it was genuinely pretty surreal watching tokens burn while lifting weights”).

Overall Impression

This video stands out as a raw, enthusiastic testimony to the practical power of AI coding agents, the evolving nature of software engineering, and the exciting yet uncertain future ahead.