Ex-GitHub CEO is Building Git, But for AI Agents

Today’s top AI Highlights:

1. GitHub was for humans. Entire is built for agents

2. Run AI-generated code safely in < 1 microsecond

3. Claude Cowork arrives on Windows

4. AI agents violate ethics 50% of the time under KPI pressure

5. Watch Claude spin up a professional Excalidraw diagram

& so much more!

Read time: 3 mins

AI Tutorial

Build an AI VC Due Diligence Agent Team with Google ADK & Gemini 3

Evaluating startup investments requires hours of research across multiple domains - company analysis, market research, financial modeling, and risk assessment. This setup literally automates this entire workflow with AI agents that work together like a real investment team.

In this tutorial, you'll build an AI Due Diligence Agent Team using Google's Agent Development Kit (ADK) and Gemini 3 models, and Nano Banana.

This 7-agent team researches any startup (from early-stage unknowns to well-funded companies), analyzes the market, builds financial projections, assesses risks, and generates professional investment reports - all autonomously with seamless handoffs and a sophisticated analysis with reports.

We share hands-on tutorials like this every week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.

Don’t forget to share this newsletter on your social channels and tag Unwind AI (X, LinkedIn, Threads) to support us!

Latest Developments

GitHub Was for Humans. Entire Is Built for Agents 📋🤖

The man who ran GitHub for four years just raised $60 million to build what comes after GitHub.

Thomas Dohmke, former GitHub CEO and the person who scaled Copilot to 20 million users, has launched Entire, an open-source developer platform designed from the ground up for agent-human collaboration.

Entire is building a three-layer stack: a Git-compatible database, a semantic reasoning layer for multi-agent coordination, and an AI-native UI that rethinks the entire software development lifecycle. The first product shipping today is Checkpoints, an open-source CLI tool that automatically captures every AI agent session — prompts, reasoning, transcripts — and pairs it with the code on every git push. It currently works with Claude Code and Gemini CLI, storing all session metadata on a separate git branch so your commits stay clean.

Key Highlights:

1. Context as a First-Class Primitive - Instead of code being a black box of agent output, Checkpoints makes the intent behind every change searchable. Developers can trace exactly what prompt and reasoning led to any piece of code.

2. Checkpoints CLI - Install via Homebrew or Go, run entire enable in any repo, and every agent interaction gets versioned alongside your code. You can rewind to any checkpoint, resume previous sessions, or search across session history to understand why code was written a certain way.

3. Agent-Agnostic by Design - Works with Claude Code and Gemini CLI out of the box, with the architecture built to support any agent. The tool hooks into your existing git workflow with zero changes to how you already code.

4. Clean Separation of Concerns - All session metadata lives on a separate branch, keeping your main branch and commit history completely untouched. Developers can query agent reasoning without polluting their codebase.

Better prompts. Better AI output.

AI gets smarter when your input is complete. Wispr Flow helps you think out loud and capture full context by voice, then turns that speech into a clean, structured prompt you can paste into ChatGPT, Claude, or any assistant. No more chopping up thoughts into typed paragraphs. Preserve constraints, examples, edge cases, and tone by speaking them once. The result is faster iteration, more precise outputs, and less time re-prompting. Try Wispr Flow for AI or see a 30-second demo.

Start flowing free

Run AI-generated Code Safely in < 1 Microsecond 📦🕛

Every team building AI agents hits the same wall: the agent needs to run Python code, but you can't just exec() arbitrary code and hope for the best. Pydantic just solved this with Monty, a secure Python interpreter written entirely in Rust.

Monty starts in microseconds, not seconds. By default, it blocks all filesystem access and all network calls. Your agent's code runs in a genuine sandbox where it literally cannot touch anything you haven't explicitly allowed. The interpreter also supports snapshotable state, meaning you can freeze an execution mid-run, save it, and resume later. That's huge for long-running agent workflows where you want checkpoints.

Supports Python, JavaScript/TypeScript, and Rust bindings, so you can embed it wherever your agent stack lives.

Key Highlights:

1. Strict Sandboxing by Default - Monty completely blocks access to the host filesystem, environment variables, and network. The agent can only call functions you explicitly register, giving you full control over what code can do.

2. Execution Snapshotting - You can pause execution at any external function call, serialize the entire interpreter state to bytes with dump(), and resume later with load() — great for durable workflows or handing off between processes.

3. Built-in Resource Limits - Monty tracks memory usage, allocations, stack depth, and execution time, and can cancel execution if any preset limit is exceeded. No runaway loops or memory blowups in production.

4. Type Checking Included - Monty supports full modern Python type hints and ships with ty for static type checking, all in a single binary, so you can validate agent-generated code before running it.

Quick Bites

Claude Cowork arrives on Windows
Anthropic brought Cowork to Windows with full feature parity: file access, multi-step tasks, plugins, and MCP connectors. Everything is folder-scoped, sidestepping the enterprise permission nightmare that's kept Microsoft Copilot's conversion rate at a painful 3.3%. If you're on Windows, this is Claude working directly with your files.

AI agents violate ethics 50% of the time under KPI pressure
New research shows that when AI agents are given aggressive performance targets, they cut ethical corners 30-50% of the time. The agents weren't instructed to cheat. They optimized for the metric and found that bending rules was efficient. So much like a human!

Qwen-Image-2.0: Open-source image gen with real typography
Alibaba's latest image model finally solves the "AI can't render text" problem. Qwen-Image-2.0 generates images with accurate typography, proper infographic layouts, and structured text. Free on HuggingFace. If you've been waiting for AI image gen that handles text without garbling it, this is the one to try.

Mastra Observational Memory hits 94.9% SOTA
Mastra's new approach to agent memory is outperforming RAG and knowledge graphs. Instead of chunking and embedding everything, it maintains an evolving text blob that updates as the agent learns. Simple concept, state-of-the-art results. Worth studying if you're building agents with persistent memory.

Tools of the Trade

1. Excalidraw MCP Server - Excalidraw’s official MCP App. Claude can create and edit diagrams natively with the entire UI coming in the chat. Search "Excalidraw" in Claude Connectors, one click to install.

2. Vibe Companion - Run Claude Code from your browser or phone. Stan Girard found a hidden --sdk-url flag in Claude Code's binary that enables WebSocket mode. Same $200/mo subscription, way more access points.

3. Airweave Agent Skills - 50+ app integrations for your agent via one npx command. Slack, Notion, Linear, GitHub, and more. Each integration becomes a tool your agent can call.

4. Awesome LLM Apps - A curated collection of LLM apps with RAG, AI Agents, multi-agent teams, MCP, voice agents, and more. The apps use models from OpenAI, Anthropic, Google, and open-source models like DeepSeek, Qwen, and Llama that you can run locally on your computer.
   (Now accepting GitHub sponsorships)

Hot Takes

1. The fastest way to validate a product:

   DON'T build it

   -Make a landing page

   -Describe the problem it solves

   -Add a "join waitlist" button

   If nobody signs up, you just saved yourself 3 months

   ~ Layton Gott
   
   

2. Vibe coding is here. Vibe research is next.

   ~ Naval Ravikant

That’s all for today! See you tomorrow with more such AI-filled content.

Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!

Unwind AI - X | LinkedIn | Threads

Awesome LLM Apps | Sponsor Us

PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉