A new paradigm in software development, termed "Agentic Engineering," is emerging, focused on building software through the use of advanced coding agents. This approach leverages tools like Claude Code and OpenAI Codex, which possess the defining characteristic of being able to not only generate code but also execute, test, and iterate upon it. This capability allows them to function with a degree of independence, reducing the need for constant, turn-by-turn guidance from a human supervisor.
A Tool for Professional Amplification
Agentic Engineering is positioned not as a replacement for developers, but as a powerful tool for professional software engineers. The primary goal is to enable practitioners to improve and accelerate their work by amplifying their existing expertise. By offloading the generation of boilerplate or initial code drafts to an agent, engineers can focus on higher-level system design, architecture, and the more nuanced aspects of software development.
The 'Agentic Engineering Patterns' Project
To formalize the practices within this evolving field, technologist Simon Willison has initiated a project titled 'Agentic Engineering Patterns'. The project's mission is to collect and document the coding practices and design patterns that yield the best results from this new generation of coding agents. The structure of the project is consciously inspired by the seminal 1994 book, 'Design Patterns: Elements of Reusable Object-Oriented Software', suggesting a goal of establishing a foundational vocabulary and set of best practices for this new domain. A software engineer at a futuristic workstation, collaborating with a holographic AI agent. Lines of code, test results, and system architecture diagrams flow between them, illustrating a seamless partnership in building complex software.
Initial Patterns and Core Principles
The project launched with two initial chapters that address fundamental shifts in the development process. The first, titled 'Writing code is cheap now,' confronts the central challenge and opportunity of agentic engineering: the cost to produce initial working code has plummeted to near zero. This chapter explores how such a dramatic economic shift impacts developers' intuitions about work, both individually and within teams. The second chapter, 'Red/green TDD,' advocates for a classic software development practice as a modern agent-guiding technique. It describes how test-first development, or Test-Driven Development (TDD), provides the necessary constraints and feedback loops to help coding agents write more succinct and, crucially, more reliable code with minimal extra prompting from the supervising engineer.
Publication Format and Development Philosophy
' A guide is defined as a collection of chapters, where each chapter is an updatable blog post with a less prominent date, designed to evolve over time rather than remain static. This format reflects the dynamic nature of the field. Adhering to a strong personal policy, Willison has stated that no AI-generated writing will be published under his name for this project. a strong personal policy, Willison has stated that no AI-generated writing will be published under his name for this project. 6 agent, accessed via a mobile device. tice, the very implementation code for the 'guide' and 'chapter' models on Willison's website was reportedly written almost entirely by the Claude Opus 4.6 agent, accessed via a mobile device. - **Definition:** Agentic Engineering involves using AI coding agents that can independently generate, execute, test, and iterate on software code. - **Audience:** The practice is intended for professional software engineers to amplify their skills and increase productivity, not to replace them. - **Formalization:** The 'Agentic Engineering Patterns' project aims to document best practices for this new field, taking inspiration from the classic 'Design Patterns' software engineering text. - **Core Principles:** Key concepts include adapting workflows to the near-zero cost of initial code generation and using Test-Driven Development (TDD) to guide agents toward producing more reliable and succinct code.
Defining Agentic Engineering
A new paradigm in software development, termed "Agentic Engineering," is emerging, focused on building software through the use of advanced coding agents. This approach leverages tools like Claude Code and OpenAI Codex, which possess the defining characteristic of being able to not only generate code but also execute, test, and iterate upon it. This capability allows them to function with a degree of independence, reducing the need for constant, turn-by-turn guidance from a human supervisor.
A Tool for Professional Amplification
Agentic Engineering is positioned not as a replacement for developers, but as a powerful tool for professional software engineers. The primary goal is to enable practitioners to improve and accelerate their work by amplifying their existing expertise. By offloading the generation of boilerplate or initial code drafts to an agent, engineers can focus on higher-level system design, architecture, and the more nuanced aspects of software development.
The 'Agentic Engineering Patterns' Project
To formalize the practices within this evolving field, technologist Simon Willison has initiated a project titled 'Agentic Engineering Patterns'. The project's mission is to collect and document the coding practices and design patterns that yield the best results from this new generation of coding agents. The structure of the project is consciously inspired by the seminal 1994 book, 'Design Patterns: Elements of Reusable Object-Oriented Software', suggesting a goal of establishing a foundational vocabulary and set of best practices for this new domain.
A software engineer at a futuristic workstation, collaborating with a holographic AI agent. Lines of code, test results, and system architecture diagrams flow between them, illustrating a seamless partnership in building complex software.
Initial Patterns and Core Principles
The project launched with two initial chapters that address fundamental shifts in the development process. The first, titled 'Writing code is cheap now,' confronts the central challenge and opportunity of agentic engineering: the cost to produce initial working code has plummeted to near zero. This chapter explores how such a dramatic economic shift impacts developers' intuitions about work, both individually and within teams.
The second chapter, 'Red/green TDD,' advocates for a classic software development practice as a modern agent-guiding technique. It describes how test-first development, or Test-Driven Development (TDD), provides the necessary constraints and feedback loops to help coding agents write more succinct and, crucially, more reliable code with minimal extra prompting from the supervising engineer.
Publication Format and Development Philosophy
' A guide is defined as a collection of chapters, where each chapter is an updatable blog post with a less prominent date, designed to evolve over time rather than remain static. This format reflects the dynamic nature of the field. Adhering to a strong personal policy, Willison has stated that no AI-generated writing will be published under his name for this project. a strong personal policy, Willison has stated that no AI-generated writing will be published under his name for this project.
6 agent, accessed via a mobile device. tice, the very implementation code for the 'guide' and 'chapter' models on Willison's website was reportedly written almost entirely by the Claude Opus 4.6 agent, accessed via a mobile device.
Definition: Agentic Engineering involves using AI coding agents that can independently generate, execute, test, and iterate on software code.Audience: The practice is intended for professional software engineers to amplify their skills and increase productivity, not to replace them.Formalization: The 'Agentic Engineering Patterns' project aims to document best practices for this new field, taking inspiration from the classic 'Design Patterns' software engineering text.Core Principles: Key concepts include adapting workflows to the near-zero cost of initial code generation and using Test-Driven Development (TDD) to guide agents toward producing more reliable and succinct code.