A New Way to Work with AI? Anthropic Completely Reinvents Source Control

Imagine this: you wake up, turn on your computer, and three AI developers are already helping with your software project. One is implementing a new feature, another is fixing bugs, and the third is testing the new code in real-time to ensure it works properly.

In the past, developers had to perform all these processes sequentially by themselves or sweat over merging the results of multiple tasks. But now, development methods are changing completely. This is because Anthropic has unveiled a new source control system and tools tailored for the ‘Agent Era.’

Why is this important?

Software development has evolved under the assumption that it is a ‘human-centric, sequential task.’ However, the situation has changed as AI agents (AI programs that perform specific tasks autonomously) have begun participating in coding. According to Anthropic’s research, as agentic coding becomes universal, the fundamental premises of traditional software development are rapidly evolving Source 1.

These changes go beyond simply increasing coding speed. As multiple AI agents perform different tasks simultaneously, traditional source control systems (systems that manage a developer’s code change history) have faced bottlenecks. The environment newly proposed by Anthropic encourages developers to act as ‘orchestrators’ who direct the work of multiple agents in real-time, drastically reducing the frequent context-switching (a phenomenon where concentration drops when moving between tasks) that occurs in this process Source 2 Source 5.

Understanding It Simply: From Git to Scion

The traditional source control system, ‘git,’ is actually optimized for managing the process of one human modifying and saving files. It is similar to several people taking turns writing in one book. But what happens if multiple AI agents modify different parts simultaneously?

Anthropic has developed a new system called ‘Scion’ Source 1.

To put it simply, if traditional git was a ‘manuscript’ for one writer, Scion is a ‘modern collaborative platform’ where dozens of writers work in their own rooms simultaneously, and a conductor moves between those rooms to combine the content in real-time. This system was designed from the ground up to adapt traditional source control principles to the scale of agents Source 1.

Current Status: The Evolution of Claude Code

Anthropic has fully redesigned its coding agent tool, the ‘Claude Code’ desktop app, to embody this philosophy Source 2.

Claude Code is no longer just a command-line tool. It now changes the developer’s workspace by including the following features Source 5:

• Session Sidebar: Allows you to see the status of multiple running AI agents at a glance.
• Drag and Drop Layout: You can freely arrange desired work windows to manage the agent’s workflow.
• Integrated Terminal: You can handle code modifications, test execution, and committing (saving) changes in one place without needing to switch between separate windows Source 5 Source 7.

Through these updates, AI agents can handle the tasks of directly reading, modifying, testing, and safely reflecting code changes in parallel sessions Source 7.

What will happen in the future?

This change by Anthropic is not just a facelift of a tool. It has set a new standard for how developers should handle AI. In the future, more AI services will adopt ‘standard protocols’ (MCP, Model Context Protocol, a common language used by different AI tools to exchange information) to connect with different agents Source 14.

The future we need to watch depends not only on AI becoming smarter, but on “how efficiently AIs can cooperate without conflict, and how efficiently human conductors can control them.” Just as source control systems have embraced the AI era, all development environments around us will transform into an era where ‘agents are the default.’

References

1. Redesigning source control at Anthropic (https://www.anthropic.com/research/redesigning-source-control)
2. Claude Code Desktop Redesign: Parallel Agents (2026) (https://www.eigent.ai/blog/claude-code-desktop-redesign)

3. Projects redesigned: from folder to conversation

   Claude by Anthropic (https://claude.com/blog/projects-redesigned)

4. Anthropic Turns Claude Code Projects Into a Parallel AI… - KingyAI (https://kingy.ai/news/anthropic-claude-code-projects-ai-agents/)
5. Anthropic redesigns Claude Code desktop app to better handle… (https://completeaitraining.com/news/anthropic-redesigns-claude-code-desktop-app-to-better/)
6. Anthropic redesigns Claude Projects as multi-agent coordinator (https://www.brocker.org/anthropic-redesigns-claude-projects-multi-agent-coordinator-claude-code)
7. Anthropic redesigned Claude Code for desktop. tl;dr This is a serious… (https://www.linkedin.com/posts/abhijeet-pachpute_anthropic-redesigned-claude-code-for-desktop-activity-7450256887431843840-lelR)
8. Google joins OpenAI in adopting Anthropic’s protocol for connecting AI agents - ZDNET (https://www.zdnet.com/article/google-joins-openai-in-adopting-anthropics-protocol-for-connecting-ai-agents-why-it-matters/)