Anthropic has released a behind-the-scenes look at the development of Claude Code, revealing that the artificial intelligence coding assistant began as an experimental internal project years before becoming one of the company’s flagship developer tools.

In a published oral history featuring engineers, researchers and executives, Anthropic traced Claude Code’s origins to 2022, when the company began exploring autonomous software engineering as a pathway toward more capable AI systems. Early prototypes struggled with even basic coding tasks, but researchers continued refining the models and the infrastructure needed to let them execute code independently.

One of those prototypes, an internal command-line tool called “clide,” allowed employees to interact with Claude directly from the terminal. Although described by developers as “wonky” and ahead of its time, the tool convinced engineers that AI-assisted software development could become practical.

Momentum accelerated in 2024 after Anthropic formed its Labs team, which focused on building what eventually became Claude Code. Engineers said the project advanced rapidly as model capabilities improved, allowing the assistant not only to write code but also search files, execute commands and fix bugs with minimal human intervention.

Claude Code launched as a research preview in February 2025 before gaining wider adoption following the release of Anthropic’s Claude 4 models and subscription plans later that year. According to team members, those improvements transformed the product from an experimental tool into one capable of handling a significant share of software development work.

Some Anthropic engineers said they now rely on Claude Code to write most, or even all, of their code, while others use multiple Claude agents simultaneously to perform research, update documents and complete development tasks in parallel.

Looking ahead, Anthropic researchers believe software engineering will increasingly shift from writing code manually to managing AI agents capable of completing complex tasks autonomously. They argue that such systems could eventually accelerate scientific research, engineering and other fields beyond software development.