Welcome back. The AI model wars have a new heavyweight. Google’s Gemini 3 Pro has topped leaderboards for just $2/million tokens, finally making "vibe coding" a reality. And that wasn’t the only surprise Google had in store for devs.

Also: Tutorial to start with Google’s agentic IDE, custom prompts for coding with Gemini 3 and 8 skills every engineering manager must have.

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.

Google launches Gemini 3 with breakthrough developer tools: The search giant dropped Gemini 3 Pro yesterday, and it's redefining the developer workflow. Topping several key benchmarks, the model makes vibe coding a reality: you simply describe the vision, and Gemini builds it.

But the bigger announcement might be Antigravity, a next-generation agentic IDE where autonomous agents take over your grunt work, handling everything from debugging to browser testing across your environment, transforming engineers from coders to project architects. See the tutorial and custom prompts for Antigravity in the resources section below.

The early verdict is in, and experts are calling Gemini 3 the real deal. AI heavyweight Andrej Karpathy has dubbed it a "clearly Tier 1 LLM" with "solid daily driver potential," praising its vibe coding capabilities and urging users to ignore the gaming of public benchmarks and just talk to the model. That sentiment is echoed by developer Theo Browne, who found the model genuinely impressive, noting it successfully one-shot a complex game dev task that completely stumped other top models.

Warp just launched its biggest Agents update yet, propelling it to #1 on Terminal-Bench and #5 on SWE-bench Verified, ahead of Claude Code and Codex CLI.

Warp’s #1 Agent now includes:

✅ Full terminal use: agents can interact with long-running commands like servers, debuggers, and more

✅ Supercharged planning: agents are steerable — you can review and edit plans as they work

✅ Full lifecycle: planning → coding → deployment

Try Warp’s Agents for yourself

Eight skills every engineering manager needs to survive industry fads: Your management style might already be outdated. In this blog, engineering leader Will Larson explains how engineering leadership expectations flip every few years: from hands-off coordinators during ZIRP to today's demand for technical depth.

How to triple engineering team engagement with better 1:1s: New research reveals the secret sauce isn't status updates — it's the 70/30 listening rule and psychological safety. Teams that shifted from project check-ins to career-focused conversations saw dramatic improvements in retention, code quality, and conflict resolution.

Why AI agents will force engineering leaders to rethink vendor relationships: Box CEO Aaron Levie says AI agents transform vendors from tool suppliers into workforce providers. For engineering leaders, this means evaluating vendors like you'd hire senior engineers — prioritizing domain expertise and performance metrics over features.

Master Google Antigravity IDE: This tutorial walks through building a flight tracker from scratch. You’ll learn to command autonomous agents that code, research, and test in parallel.

How to build your first recommendation system (easy): Devs can build their own recommendation system using this practical PyTorch tutorial. This walks through the step-by-step process of training and serving a collaborative filtering model to serve users content.

n8n course for beginners: This course teaches you how to use n8n by building four practical projects. You'll learn why standards like REST and OAuth2 are key to automation, how to integrate AI Agents into your workflows, and lots more.

SciAgent: In this paper researchers propose SciAgent, a unified system using specialized sub-agents for cross-disciplinary scientific problem-solving. The framework surpasses human gold-medalist scores on international Olympiads, enabling engineers to build adaptable AI for complex multi-domain reasoning.

The era of agentic organization: This paper introduces a thinking protocol called AsyncThink — it enables AI agents to solve problems collaboratively through concurrent thinking structures. The method reduces inference latency and improves mathematical reasoning accuracy, offering engineers faster and more accurate multi-agent systems.

Solving a Million-Step LLM Task with Zero Errors: This paper describes MAKER, the first system that successfully solves a task with over one million LLM steps with zero errors, and, in principle, scales far beyond this level. The approach relies on an extreme decomposition of a task into subtasks, each of which can be tackled by focused microagents.

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

Until next time — The Code team