Welcome back. Chinese AI labs are quietly minting billions in revenue. According to a top VC, four of China's five largest private labs already rank among the top 25 worldwide for revenue. Adding to the momentum, one of them just dropped a model that outshines Fable 5 on frontend tasks. 

Also: How to pick the right GPT-5.6 model, a software engineer’s guide to AI stack, and the hiring truth behind a viral Amazon story.

Moonshot’s new model beats Fable 5 for frontend: The Chinese AI lab just released Kimi K3. It’s a massive 2.8 trillion parameter model with native vision and a 1 million token context window. The lab claims it only trails Claude Fable 5 and GPT 5.6 Sol, meaning it beats every other model they tested. It even beats Fable 5 on the Frontend Code Arena, where human judges compare the designs side-by-side. The full weights will be available on July 27.

Claude Code now reviews your PRs as hard as you want: The AI lab just rolled out new effort levels for /code-review. Previously, the command ran one fixed prompt no matter what. Now, each level runs its own review. Low effort runs a fast single pass that you can use before every push. High effort spins up sub-agents to verify every single finding. Anthropic says even the lowest tier outperforms rival tools. Update Claude Code and try it on your next PR.

DoorDash lets your AI agents order food: The food delivery giant just opened a limited beta of dd-cli. This command-line tool lets agents search stores, find deals, and handle checkout. You can now integrate food ordering directly into their workflows instead of using the app. Early access is available via waitlist for macOS developers in the US and Canada.

AI coding tools can speed up writing code, but for most teams, the real bottleneck is finding the right context.

This guide breaks down the four categories of AI tools for engineering teams, explains the two-layer model leading organizations are adopting, and offers a practical framework for evaluating tools based on context depth, explainability, security, governance, and workflow fit.

Get the guide to build an AI stack that actually helps your team ship faster and more reliably.

The lone sprinter problem. AI adoption follows a familiar pattern at most companies. One engineer might be 10x'ing their output with Claude Code. Meanwhile, the rest of the org hasn't caught up yet. Claude Code creator Boris Cherny hears this story every single day. So he mapped out the exact roadmap teams take to close that gap.

Four steps to AI native adoption. His map tracks how the engineer's role changes:

Bottlenecks not budgets. The real question is what actually scales a team from one stage to the next. Simply buying more tokens won't cut it. The biggest challenge is domain knowledge that lives in your engineer’s head. To overcome this, Cherny recommends documenting everything and building markdown files for your agents. It ensures that technical knowledge doesn't just sit in silos but becomes an active asset for the entire team.

Every line earns its place. Most teams struggle when it comes to documenting domain expertise.The natural instinct is to over-document, but models start losing the plot after a few hundred lines. HumanLayer's guide shows how to keep a CLAUDE.md inside that budget, and Anthropic’s guide on writing effective skills will help you document deeper workflows that your agents can use when needed.

You pay for every token your agent burns. When it returns code that doesn't fit your system, you prompt again. And again. That back and forth is because your agents are still missing critical context. Join live and for free on July 23 to learn more.

Register now.

The complete AI Coding workflow: In this tutorial, a senior engineer shows you how to set up an AI-native environment for professional work. You’ll learn how to install and configure skills using the CLI. You’ll also learn how to organize your projects for AI agents and run through a standard workflow.

CodeMote: This tool connects your iPhone to your machine to run agent sessions. It features a live terminal and push alerts for fast approvals. The connection is private and fully encrypted. 

Claude HUD (26.5K ⭐): A Claude Code plugin that gives you a clear view of your context usage and active tools. You can also track running agents and check your progress on tasks.

Prompting guidance for GPT-5.6 Sol (by OpenAI): Upgrading to GPT-5.6 requires a fresh approach to maximize token efficiency and new features. This cookbook shows you how to use leaner prompts and Pro mode to boost performance and lower your costs.

Code merges faster than tests can be written. Once a PR is merged, it's rarely touched again. The Cursor engineering team solves this with a scheduled agent. Their blog explains how they use morning automation to review recent merges and fill in missing tests.

You can set this up with a single prompt in Cursor 3.8 or later:

Cursor builds the schedule and tools from that description. Every morning, you can review a small test-only PR.

P.S. Get 50+ AI coding hacks for Claude Code, Cursor, and Codex here.

Thinking Machines recently dropped Inkling. It's a massive, open-weights model with 975B parameters. Now, you can run it locally. See how.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 300K+ engineers and 150K+ followers on socials. Get in touch.

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team