Vercel ships tool that grades sites for AI agents

Welcome back. A few months back, Cloudflare research revealed that agents now drive over half of all web traffic. Your next wave of customers won't be human. They're agents. If your site isn't optimized for them, you might already be falling behind. Vercel just dropped a tool to check how agent-ready your frontend is. Read details below.

Also: How an engineer’s side hustle banked $45K in two months, a one-shot project to build your AI engineering skills, and how to cut AI costs with token engineering. 

Vercel’s new tool grades websites for agents: The cloud hosting platform just dropped Is Agentic, a new tool that scores how easily AI agents can navigate and use a website. Every scan is run by Ora, an agent-experience research firm. It runs over 100 checks to see how agents navigate your site. You get a score, one-click fixes, and a CLI for agents to run scripts directly. It's free, with no API key or billing. You can run it here. 

Shopify CEO builds an open-source Git server: The e-commerce platform's founder just launched Walgit, a single-binary Rust server that turns cheap cloud storage into a code host with no database behind it. He built it over a weekend after reading Cursor's write-up on scaling Git. It clones giant repos fast by serving them as plain static files. It's MIT-licensed with a JSON API, so you can self-host cheaply and build your own tooling on top. You can try it here.

Anthropic expands access to Mythos for hunting vulnerabilities: The AI lab just put Mythos 5, once limited to a chosen few, inside a security scan that enterprise teams can turn on themselves. You can simply point it at your GitHub repo to surface flaws tagged by type, confidence, and severity. You maintain full control by manually approving every fix, while it scans the bill as standard token usage on your existing plan. When you open a patch in Claude Code, it uses the models your enterprise team already uses. See how it works. 

AI is writing more and more code than ever before in history. 

So why are you still reviewing PRs like it’s 2010—alphabetical order, zero context, no high-level summary of why changes were made?

CodeRabbit’s Change Stack restructures PR reviews into layered walkthroughs, giving you…

And if you need help? Just ask the Agent Chat for context directly where you’re working.  

Try it here (early access costs nothing).

Agent burnout. A single coding agent can only do so much. If you give it too many tasks at once, it starts breaking one thing while trying to fix another. Even experienced engineers like Bob Martin (also known as Uncle Bob on the internet) have run into this problem: finding that agents struggle with messy code and technical debt just as much as humans do.

Change the numbers. During a podcast with Matt Pocock, Martin explained that he moved away from using a single agent for everything. Instead, he now routes each user story through a five-agent assembly line where every agent is responsible for a single step. Here’s his setup:

The power of context. Splitting tasks keeps the prompt short so the agent actually follows the rules. Each agent does its job and then resets. This keeps them laser-focused on one thing from start to finish.

The math checks out. One agent might finish in five minutes, but you can't trust the output. The full hour-long pipeline produces battle-tested code that would've taken a human half a day. While Martin still handles the high-level architecture and module design, the grunt work is now fully automated.

PS: Want to see it in action? He’s got two-, four-, and six-agent versions ready to go here.

Sonar Vortex operates inside your AI coding agent’s inner reasoning loop, supplying architectural context before code is written, then verifying the output in real time. In testing, software defects dropped 92% and token consumption declined 30+%.

Build safer code while spending less on LLM calls.

How to build an autonomous AI coding system: This tutorial is all about agentic AI software engineering. You'll learn how to build a 5-component automated pipeline that handles everything from planning and testing to shipping software straight from your PRD specs. 

Anti-slop for AI-written code (3.5K ⭐️): This plugin catches low-signal TypeScript and JavaScript patterns before they land in your codebase. You can vendor the rules into your repo, customize them to your standards, and even have your coding agent install and configure everything for you.

How to cut AI costs with token engineering (by Google): Building cheaper AI systems takes more than trimming prompts. This guide shows you how to prune and cache agent context, batch offline work, cap reasoning tokens, and route simple tasks to smaller models, so AI workloads stay fast, reliable, and cheaper to run at scale.

Codex quota resets weekly and unused credits vanish. A dev shared a /goal prompt that turns leftovers into a full UI/UX audit.

If /goal is missing, enable it:

Restart Codex, then paste:

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

Check out how one dev tried simulating a whole company using SpaceXAI’s new Grok Bot. It only took a few minutes, but a single catch brought everything to a halt mid-task. 

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