Welcome back. If you’re running MongoDB, your morning just got a lot more urgent. Hackers are currently exploiting a critical flaw that bypasses login credentials on tens of thousands of servers worldwide.

Once you’ve patched your systems, stick around for some better news: Alibaba might have just made Midjourney obsolete with their new open-source model that gets character consistency right.

Also: Anthropic’s official Claude Code course and pro hacks for GitHub Copilot users.

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.

Alibaba upgrades its open-source image editing model: The Chinese tech giant just dropped Qwen-Image-Edit-2511, a major upgrade to its AI image editor that tackles the biggest pain point in the space: consistency. The new model dramatically reduces image drift, preserves character identity across edits, and adds built-in LoRA support — no extra tuning required. It also features improved geometric reasoning and better industrial design generation. You can try it now in Qwen Chat or deploy it locally via ModelScope for best results.

Critical MongoDB flaw puts passwords at risk: Hackers are actively exploiting a bug in MongoDB that lets them steal sensitive data — including passwords, user info, and API keys — without needing login credentials. Over 87,000 servers worldwide are vulnerable. CISA has flagged the issue, giving federal agencies until January 19 to patch. If you're running MongoDB, update to the latest version now to stay protected.

Google recaps its biggest AI moments of 2025, plus 40 helpful tips: The tech giant published a year-end summary of its AI accomplishments, including updates to models, products, scientific advancements, and more. Big themes of the year included conversational AI in Search, coding tools, and autonomous agents that don't just answer questions, but take action. The search giant also dropped 40 of its most helpful AI tips from 2025.

10 contrarian leadership truths every leader needs to hear: In this podcast, Rippling’s CPO reveals how deliberately understaffing projects can actually reduce office politics and waste. MacInnis explains that maintaining high intensity is vital to fighting the natural drift toward disorder in teams. He also shares how to use quality checklists to ensure consistency without stifling innovation.

Why blaming AI for bad code is a leadership red flag: Some teams are ditching AI coding tools entirely because reviewing AI output is "harder than writing it themselves." Others push 60 PRs a day without proper review. Both extremes miss the point. The real issue is accountability — engineers must treat AI-generated code as their own responsibility. Leaders should enforce this mindset: test it, understand it, own it. AI amplifies your practices, good or bad.

Why engineering leaders should capture decision traces: As AI agents take over enterprise workflows, the real competitive advantage isn't your data — it's documenting why decisions were made. The reasoning behind exceptions, approvals, and precedents currently lives in Slack threads and people's heads. Teams that build systems to capture this "context graph" will unlock agent autonomy faster than those relying on traditional systems of record.

Claude Code course (by Anthropic): In this tutorial, you’ll learn agentic coding with Claude Code to navigate, debug, and refactor codebases. You will build real world projects — including a RAG chatbot and e-commerce dashboard, and learn to use Figma and Playwright MCP servers to automate workflows and turn designs into functional web applications.

How to use Agent skills in VS Code: Learn to teach GitHub Copilot specialized tasks using Agent Skills. By creating skill.md files, you can provide targeted instructions that load only when needed, preserving your context window. You'll see practical examples ranging from framework migrations to automating command-line video editing.

How to do Spec-driven development with AI: If you are using GitHub Copilot, this tutorial teaches you to use Spec Kit to turn vague prompts into reliable software. You will learn to steer the AI with clear specifications and plans, ensuring it generates high quality code that exactly matches your intent instead of guessing.

A plan reuse mechanism for LLM-driven agent: This paper discusses the slowness of AI agents constantly creating new plans for similar tasks. It proposes a reuse method that recycles old plan structures with new details, cutting wait times by over 93% without losing accuracy.

Sophia: A persistent agent framework of artificial life: This research addresses the issue that current AI agents stop learning once deployed, staying static. It introduces "Sophia," a solution using a "System 3" layer that allows agents to set their own goals and continuously improve without human help.

Agent-R1: This framework tackles how standard training methods fail to handle the unpredictable, multi-step interactions of autonomous AI agents. It presents Agent-R1, a new framework that successfully adapts reinforcement learning for these dynamic environments, significantly outperforming traditional approaches on complex tasks.

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

Until next time — The Code team