The Tea ☕
☕ Garlic: OpenAI’s New Small Model That Beats Gemini 3 & Opus 4.5 in Coding Benchmarks
Source: Reddit (Dec 4, 2025)
OpenAI is developing a new AI model codenamed "Garlic" to compete with Google's Gemini 3 and Anthropic's Opus 4.5, particularly in coding and reasoning tasks. CEO Sam Altman declared an internal "code red" following Gemini 3's rapid success, signaling an urgent push to advance ChatGPT and regain competitive leadership.
Chief Research Officer Mark Chen informed employees that Garlic has outscored both Gemini 3 and Opus 4.5 in internal benchmarking tests. The model could debut as GPT-5.2 or GPT-5.5 by early 2026, and it incorporates efficiency breakthroughs that allow a smaller model to contain the same knowledge that previously required a larger architecture.
The development comes as Google's Gemini monthly active users climbed from 450 million in July to about 650 million by October, intensifying competitive pressure on OpenAI.
☕ IDEsaster: 30+ Critical Flaws in AI Coding Tools Enable Zero-Click Data Theft and Code Execution
Source: The Hacker News (Dec 6, 2025)
Security researcher Ari Marzouk has identified over 30 vulnerabilities in AI-powered coding assistants and IDEs, collectively named IDEsaster. These flaws affect popular tools, including Cursor, Windsurf, Kiro.dev, GitHub Copilot, Zed.dev, Roo Code, Junie, and Cline, with 24 assigned CVE identifiers.
The attack chains exploit three key weaknesses: bypassing LLM guardrails through prompt injection, executing auto-approved tool calls without user interaction, and weaponizing legitimate IDE features to breach security boundaries. The vulnerabilities enable data exfiltration and remote code execution via prompt injection combined with legitimate IDE features.
The researcher emphasized that all AI IDEs tested effectively ignore their base software in threat modeling, treating features as inherently safe despite AI agents being able to weaponize them. Marzouk recommends using AI IDEs only with trusted projects, monitoring MCP servers for changes, and manually reviewing sources for hidden instructions.
☕ Anthropic Brings Claude Code to Slack, Enabling Full Coding Workflows in Chat
Source: TechCrunch (Dec 8, 2025)
Anthropic is launching Claude Code in Slack as a beta research preview, enabling developers to delegate complete coding tasks directly from chat threads. Previously, developers could only get lightweight coding help via Claude in Slack, like writing snippets, debugging, and explanations. Now teams can tag @Claude to initiate full coding sessions using Slack context like bug reports or feature requests.
Claude analyzes recent messages to determine the right repository, posts progress updates in threads, and shares links to review work and open pull requests. This reflects a broader shift where AI coding assistants are moving from traditional IDEs into collaboration platforms where teams already work.
For Slack, positioning itself as an "agentic hub" where AI meets workplace context creates a strategic advantage. However, the integration raises concerns about code security and IP protection, as outages or rate limits in either Slack or Claude's API could disrupt development workflows that teams previously controlled locally
🤖 Prompt of the Week (#POTW)
The Auto-Responder That Actually Helps
When to use: You need an out-of-office message that doesn't just say "I'm gone" but actually solves problems.I'm going offline from [START DATE] to [END DATE]. Help me create an out-of-office auto-responder that: - Sets clear expectations about response time - Provides alternative resources for common requests - Directs urgent matters to [BACKUP PERSON] - Maintains my professional tone - Includes specific guidance for these typical inquiries: [LIST YOUR 3 MOST COMMON REQUEST TYPES] My usual communication style is: [warm/direct/formal/conversational]. The auto-responder should sound like me, not a corporate robot.
Why it works: Most out-of-office messages are useless. This one anticipates problems and routes people to solutions before they become your emergency.
🛠️ The Toolbox
Your weekly must-try AI tools & resources
BRAND•E: Built for agencies and multi-brand companies who need to scale content production without writing a single prompt...while keeping each brand's authentic voice intact across every piece of content.
→ Try BRAND•E FREEMake.com: Make gives you real-time visibility and control of all your AI automation and agents so you can scale with confidence.
→ Get Started FREELoom: Loom is a video messaging tool that lets you record your screen, camera, or both in seconds.
→ Try Loom FREE
💬 Community Corner
The "Set It and Forget It" Sprint
Your mission: Build one complete automation that runs while you're offline. Choose your fighter:
Email automation: Set up a sequence that nurtures leads, follows up on proposals, or sends resources to common questions
Social media scheduler: Load 2 weeks of content with strategic posting times and conversation-ready copy
Task handoff system: Document one recurring responsibility so thoroughly that someone else could own it completely
Success looks like: You test your automation, it works perfectly, and you feel a small surge of relief knowing this won't be your problem during time off.
Tools to try: Zapier, Buffer/Later/Hootsuite, Notion + Loom for documentation
Your AI Bestie who's grateful you're here
Catch you next Wednesday,

![Holiday Mode: Setting Up AI to Work While You Don't [12/10/25]](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,quality=80,format=auto,onerror=redirect/uploads/asset/file/239532c1-ea9a-4754-a77e-03e1590f0590/AdobeStock_1777052873.jpeg)