Former GitHub CEO's Entire Redefines Git for the Agent Era

Can you explain why an AI agent wrote code a certain way three months from now?

Git records what changed, but not why those decisions were made. Entire, co-founded by former GitHub CEO Nat Friedman, tackles this problem head-on.

The Founders

Nat Friedman served as GitHub CEO from 2018 to 2021 and was instrumental in pushing Copilot during its early stages. Co-founder Daniel Gross is a former Y Combinator partner and led Apple's machine learning team. The pair previously ran AI Grant together, a program funding AI researchers.

Friedman wrote in the launch blog post:

"Git was built in 2005 for Linux kernel development. It's a tool from an era when humans wrote code. Today, more than half of code is written by agents. It's time for the tools to evolve."

Core Feature — Checkpoints

Entire Checkpoints is a layer on top of existing Git. It stores not just code changes but the AI agent's decision-making context—prompts, intermediate reasoning, and alternative comparisons.

1. Context Checkpoints

Bundles code diffs with decision records. Where a normal Git commit records "line 23 of file A changed," an Entire checkpoint records "the agent compared 3 approaches and chose method B based on performance and maintainability considerations."

Metadata stored:

Prompt history: Complete prompt chain sent to the agent
Reasoning traces: Intermediate chain-of-thought
Alternative analysis: Approaches considered but rejected, with reasons
Context window: Files and documents the agent referenced

2. Agent Replay

Reproduce an agent's thought process at any point in time. When debugging, instead of asking "why does this code look like this?"—you can replay the exact information the agent saw and the reasoning it went through.

3. Branch Comparison

Visually compare alternatives the agent considered. Not just code diffs, but side-by-side analysis of each alternative's pros and cons.

4. Team Collaboration

When multiple agents work simultaneously, track how each agent's decisions influenced others. If Agent A modified a database schema and Agent B adjusted the API accordingly, view the entire chain on a single timeline.

Technical Architecture

Entire does not replace Git. It creates an .entire directory alongside the existing .git directory and maintains bidirectional sync.

Aspect	Git	Entire
Tracks	File changes	File changes + decision context
Unit	Commit	Checkpoint
Metadata	Message, author, timestamp	+ prompts, reasoning, alternatives, references
Compatibility	—	100% compatible with existing Git
Storage	Local + remote	Local + remote + optional cloud

The entire CLI wraps Git commands directly. entire commit internally runs git commit while storing additional metadata in .entire/.

Why This Matters Now

According to GitHub's 2025 Octoverse report, AI-contributed code in Copilot-enabled projects averaged 46%, up from 35% in 2024. With agent coding tools like Cursor, Windsurf, and Claude Code, this ratio is projected to exceed 60% in 2026.

Auditability Requirements

Regulated industries (finance, healthcare, defense) need AI-written code to provide rationale for implementation decisions. The EU AI Act requires decision traceability for high-risk AI systems. Git commit messages alone cannot meet this requirement.

Multi-Agent Collaboration

Scenarios where multiple agents work on the same project simultaneously are becoming common. When Devin, OpenHands, and Claude Code each modify different modules, understanding how their decisions conflict or depend on each other requires information beyond commit messages.

Competitive Landscape

Other players in the "AI agent development infrastructure" space:

Cognition (Devin): Focused on the AI agent itself—writing code is core, uses standard Git for version control
Replit Agent: Automates from env setup to deployment but lacks decision tracking
GitButler: Improves Git UX for human developers, not AI agents

Entire complements rather than competes. Whatever agent tool you use, it provides infrastructure to track that agent's decisions.

Funding

Entire raised $85M in a Series A in February 2026, led by Andreessen Horowitz's AI infrastructure fund. a16z recently published an investment thesis on AI DevTools infrastructure, stating that "the era of agents writing code demands a new class of development tools."

Version Control Through the Ages

Version control has seen paradigm shifts roughly every 15–20 years:

1972: SCCS — single-file version control
1986: CVS — multi-file, central server
2000: SVN — improved CVS, atomic commits
2005: Git — distributed version control, built by Linus Torvalds for the Linux kernel
2026: Entire — tracks AI agent decision-making

Each transition point involved a shift in who writes code. Git was built for open-source distributed collaboration; Entire is being built for human-AI collaboration.

DVC and MLflow — Existing AI Version Control

The ML world already had similar tools:

DVC (Data Version Control): Version control for datasets and models, used alongside Git
MLflow: ML experiment tracking, model registry
Weights & Biases: Experiment logging, visualization, hyperparameter tracking

These track the model training process. Entire tracks the code generation process. Different problems, but sharing the principle of "tracking AI decisions."

The AI Observability Market

Entire's broader category is "AI Observability." Existing monitoring tools like Datadog and New Relic are adding AI-specific features, while startups like LangSmith (LangChain), Arize AI, and Helicone target this space. Gartner estimates the AI Observability market will reach $9.5B by 2028. Entire's unique positioning focuses on development-time decision tracking — differentiated from other tools that focus on production monitoring.

Implications

Entire's approach goes beyond version control. In an era where agents write code, it builds the foundation for human developers to understand and verify AI decisions.

The message for developers is clear: going forward, "why it changed" matters more than "what changed." If Git was the foundation for code collaboration, Entire aims to be the foundation for human-AI collaboration.

Currently in Private Beta with a waitlist. GA is planned for Q3 2026.

Former GitHub CEO's Entire Redefines Git for the Agent Era

The Founders