Название: Code That Builds Itself: A Practical Guide to Agentic AI Development Workflows
Автор: Eric Smith
Издательство: BGB Learn
Год: 2026
Страниц: 171
Язык: английский
Формат: pdf, epub (true)
Размер: 10.1 MB

The developer's role is fundamentally shifting from writing code to directing it. Agentic AI can now read specifications, explore codebases, write implementation code, create tests, fix bugs, and submit pull requests without a developer typing a line of implementation. This practical guide is for engineers who want to master this new paradigm and stay ahead of the curve.

This book provides a framework for moving beyond autocomplete to build a development practice where you and your AI agents create software together. You will learn how to choose the right tools, manage context, and review AI-generated code effectively. It addresses the unique challenges of this new workflow, including quality debt, cognitive decline, and security. Understand the mental models needed to work with LLMs and implement prompt patterns that scale.

Whether you are using Cursor, Cline, or other agentic tools, this book will show you how to set up for success, build multi-agent workflows, and manage costs. It also covers the crucial topics of security, governance, and the cognitive debt paradox. This is the essential resource for developers who want to build software faster and more reliably by forming an effective partnership with AI agents.

Every few years, a technology arrives that changes not just how we write software but what it means to be a developer. The compiler was one such invention, taking assembly language and letting programmers think in higher-level abstractions closer to the problems they were solving, eliminating entire categories of human error. The graphical operating system was another, replacing command-line interfaces with visual metaphors and opening software development to people who would never have touched a terminal. The internet was yet another shift, turning software from a local artifact into a global system where distribution, security, and network effects became as important as core application logic. Each was met with skepticism by incumbents who had mastered the previous paradigm. Each ultimately expanded the profession, creating new categories of work and requiring skills the previous generation could not have anticipated. The pattern is consistent enough that we should recognize it when it appears again.

This is not a prediction about what might be possible in 5 years if the technology continues improving. It describes tools that exist today, available for download, used right now by developers at companies of every size to build software real customers depend on. Cursor, Cline, Claude Code, Copilot Agent Mode, Windsurf, Codex CLI, and a dozen other tools are already being used by hundreds of thousands of developers, and the adoption curves look like the early days of a platform shift. Companies like Ramp, Notion, and Rakuten have reported shipping code written primarily by AI agents, with human developers serving in review and oversight roles. Open source projects like OpenCode have accumulated hundreds of thousands of lines of code written almost entirely by autonomous agents, with quality metrics competitive with human-written code by conventional measures. The question is no longer whether this technology works reliably enough for production use. The question is whether you know how to use it well, because the difference between transformative value and frustration is almost entirely a matter of approach.

Contents: