jimmytoan/ai-coding-survival-guide

This repository is a practical guide for using AI coding assistants more thoughtfully, particularly in the context of usage-based billing where every AI model call carries a cost. The guide opens with the "Max Power" metaphor, a reference to a cartoon character who insists on doing things the hard way. The point is that many developers use AI tools in ways that produce fast output but also generate fast waste: giant context windows, expensive models for simple tasks, and long iterative loops that undo what was just generated. The core argument is that the most effective use of AI coding tools is not about generating the most code, but about controlling context, selecting the right model for the task, scoping work before handing it to an agent, and reviewing output carefully. The guide describes this as the beginning of "FinOps for AI coding agents," borrowing a cost-management discipline from cloud infrastructure and applying it to AI model usage. The repository includes a set of documents covering specific topics: context discipline (when to start a fresh chat and how to write a useful compacted summary), model selection (using cheaper models for simpler tasks and stronger models for architecture decisions and security-sensitive code), agent mode usage (when delegation makes sense and when it does not), team guidelines, cost observability, and prompt pattern examples showing good and poor formulations side by side. The guide does not promote any specific tool and is described as anti-waste rather than anti-AI. There is no runnable code in the repository. Readers consume it by browsing the docs directory, with a table in the main README pointing to each document by topic.