Claude Code: Up and Running

Foreword

For much of my recent work with generative AI, I’ve been thinking about how we can put AI into the applications we build. How can we give an application access to an LLM, tools, memory, and data? How can we build applications that aren’t merely powered by AI, but are increasingly agentic?

Meanwhile, something equally interesting has been happening in the other direction: AI has been finding its way into the process of building those applications.

That’s where Claude Code comes in.

Software developers have always relied on tools. Editors became IDEs. IDEs learned to complete code, find errors, perform refactorings, and automate increasingly sophisticated tasks. More recently, AI coding assistants began suggesting entire lines, methods, and classes as we typed.

Agentic coding takes another step. Instead of simply asking an AI to generate a snippet of code, we can give it a goal and let it work toward that goal. It can explore a codebase, determine which files are relevant, make changes, run tests, discover that something didn’t work, and try again. It can help us understand unfamiliar code, investigate bugs, review changes, and interact with other development tools. That changes the relationship between developer and tool in some interesting ways.

I’ve been developing software for a long time, and I’ve seen plenty of technologies arrive with promises that they would revolutionize the way we write code. Some did. Others proved considerably less revolutionary than the initial ...