Welcome back. AI writes code. It reviews code. But making that code secure? That was still on you — until now. Anthropic just released Claude Code Security, and the market didn't need long to react. Cybersecurity stocks dropped sharply within in an hour of the announcement, with CrowdStrike, Cloudflare, and Okta all taking a hit.

Also: How to master coding agents, how to turn OpenClaw into a dedicated AI employee, and monetize it with five wrapper ideas worth building.

Welcome to The Code. This is a 2x weekly email that cuts through the noise to help devs, engineers, and technical leaders find high-signal news, releases, and resources in 5 minutes or less. You can sign up or share this email here.

Anthropic launches AI security scanner, shaking up cybersecurity stocks: The AI lab just dropped Claude Code Security, a new tool built right into Claude Code that scans projects for vulnerabilities and suggests patches for developer review. Using Claude Opus 4.6, the team has already found over 500 bugs in live open source code that went undetected for decades. The news hit the market hard, with cybersecurity stocks tanking within an hour of the release. CrowdStrike fell 6.5 percent, Cloudflare dropped 6 percent, and Okta went down 5.7 percent.

Google unveils Gemini 3.1 Pro with a major leap in reasoning: The search giant just shipped Gemini 3.1 Pro, a smarter model designed for complex problem solving. It more than doubled the reasoning performance of Gemini 3 Pro on ARC-AGI-2, a benchmark that tests how models handle brand new logic patterns. The update brings major improvements to agentic coding, multi-step workflows, and tool use. Click here to see Gemini 3.1 in action.

Sam Altman pushes back on AI energy concerns: Speaking at an AI summit in India, OpenAI CEO Sam Altman shut down viral claims about ChatGPT's water and energy footprint. He called the water concerns "totally fake," explaining that data centers have moved on from evaporative cooling. Altman also dismissed the idea that a single query drains one and a half iPhone batteries, saying it's nowhere near that high. While he agreed that total energy use is a valid concern, he argued that AI might already be as efficient as a human on a per-query basis.

Most people don’t know how to get the right results from Conversational AI. It hallucinates. Breaks character. Drifts off-script. Fumbles edge cases. 

The problem isn’t the model; it’s your agent prompting. 

This Prompt Engineering Guide from ElevenLabs shows you 6 steps to design enterprise-grade agents with clear goals, strict guardrails, smart tool use, and built-in evaluation loops. 

The results? Agents that resolve faster, stay compliant, and improve with every interaction.

Learn how to correctly prompt your Conversational AI (before it erodes customer trust).

SaaS interfaces are losing their grip. A growing number of founders have stopped logging into their apps entirely, choosing to run their entire stack through AI agents connected to APIs. Fintool CEO Nicolas Bustamante recently shared an article outlining three forces turning every SaaS product into an API endpoint:

The real threat isn't technical. When AI agents handle software tasks, switching costs collapse because moving from HubSpot to Attio is as simple as changing an API endpoint. Companies with clean, fast, and well documented APIs will become the top choice for agent based workflows.

Every new LLM, plugin, and API widens your attack surface. All it takes is one breach, and customer trust evaporates. 

Knowledge is power: Datadog’s AI Security Best Practices eBook will teach you how to protect your app despite rapidly evolving threats. 

It lays out the common, hidden risks to prepare for, specifically how: 

Download it at no cost.

Large migrations involve making the same changes across hundreds or thousands of files. If you run Claude Code manually on every single one, you'll hit your context window. Anthropic's official documentation suggests using a bash script to trigger headless sessions across your whole file list instead.

Have Claude generate the list of files that need the change:

Save that to files.txt. Then run this loop:

The -p flag triggers headless mode, while --allowedTools locks Claude down to only editing files and making git commits.

Start by testing a few files to refine the prompt before scaling up. You can use the --output-format json flag to pipe results into a log file, making it easy to automate retries for any files that fail.

How to use built-in git worktrees in Claude Code: Anthropic just added built-in git worktree support for Claude Code. This guide walks you through using it to run AI agents in parallel without breaking anything. You'll learn how to keep tasks isolated, stay on top of your branches, and make sure you don't accidentally push AI commits straight to main. 

NanoClaw: A secure, containerized alternative to OpenClaw, runs directly on Anthropic's Agents SDK. It includes WhatsApp integration, long-term memory, and support for scheduled jobs.

Benchmarking how well agent skills work: AI agents often struggle with specific step-by-step knowledge, and until now, it was unclear if adding pre-packaged "Skills" actually helped. Testing proves human-made skills significantly boost success, but models completely fail at creating these instructions themselves.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 150K+ engineers and 100K+ 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