You can now build a lifelike agent with a face

Welcome back. You can now expect realistic AI avatars to enter your Google meetings. The search giant just dropped a model with live avatars that can think and talk like humans. Meanwhile, Meta actually launched a pendant that you can wear to talk to your agents.

Also: Meta’s Muse agent came to the rescue when a dev’s flight got delayed, 4 AI models were asked to push a man to death (one agreed), and an AI coding dictionary for beginners.

You can now build an AI avatar that can think and talk in real time: Google just dropped Gemini 3.8 Live with Live Avatar, blending real-time video with speech. You can now transform a single reference image into a fully animated avatar, though custom setups require enterprise allowlisting. Under the hood, it handles reasoning and tool calls asynchronously, so your agent fetches data mid-conversation without breaking flow. It supports 97 languages and includes complete API docs. 

Cursor puts bots on deployment and security duty: The AI startup just launched two bots to take on the hardest parts of shipping code. The Rollouts bot tracks your change from PR to production, catches broken endpoints, and can pause rollouts or open a revert PR. Meanwhile, Security Reviewer scans every update for risks, running 21% faster with higher fix acceptance rates. Both bots are live now on Teams and Enterprise.  Turn on in Automations.

Meta wants you to wear your AI agent: Mark Zuckerberg took the stage at the Connect conference to unveil the new Meta VR Glasses, a slimmed-down headset with a tethered compute puck, aimed directly at Apple's Vision Pro. He also introduced Muse Charm, a handheld pendant for calling up Meta's Muse agent, slated for the December holidays. For engineering teams, this opens a brand-new, always-on surface for agentic apps. 

Growing Postgres databases eventually hit a wall: queries slow, dashboards lag, and adding another database starts to feel inevitable.

Tiger Data keeps Postgres fast as your data grows with time-series primitives, columnar storage, and automatic partitioning. Same SQL and tooling. No pipelines or second system.

Just Postgres, built to scale with your workload.

Start building for free today. No credit card needed.

Old book, new tricks. Agents can build full-stack apps in minutes, but subtle bugs in the codebase can grind your build to a halt. Senior dev Matt Pocock kept hitting this exact hurdle, and his workaround to this wasn't a shiny new model or a clever prompt hack. But, it came straight from a classic programming book published back in 1999. He explained the idea on a podcast.

Here's the copy-ready version for you:

The magic word. These terms are basically prompt shortcuts. Say "use a tracer bullet," and the agent can pull up an entire pattern it already understands. That is much clearer than explaining every step yourself.

The readers win. Turns out, reading the classics still matters. If you know your fundamentals, you already have a library of powerful prompts that someone using plain English might never think to try. And as agents write more of the code, that advantage only gets bigger.

P.S. Ready to raid the classics for more magic words? Start with these eight programming books that shaped the field.

How to build an agentic software factory: This deep dive shows how to turn AI agents into a loop that keeps improving your software. You’ll see how agents collect issues from telemetry and user feedback, reproduce bugs, ship fixes, verify them end to end, and only then open a PR, so the workflow moves beyond “write code” into something much closer to a self-improving engineering system.

AI Engineering From Scratch (57k ⭐): A free, open-source curriculum for learning AI by actually building it. It spans 523 lessons across 20 phases, from math and deep learning to LLMs, agents, MCP, and production, with every lesson producing something reusable like a prompt, skill, agent, or MCP server.

How to get most out of Opus 5.5: Anthropic’s guide explains how to prompt the model for longer coding runs without micromanaging it. The big tips: define what “done” looks like, stop telling it to “think hard,” use CLAUDE.md to say when it should keep going or stop and split big jobs across subagents.

Hitting the five-hour session limit mid-task kills momentum. A dev shared a command Anthropic shipped quietly around v2.1.241 without ever adding it to the changelog.

One catch: this still draws from your weekly limit, so use it to finish the task in front of you, not to start new ones. Run /usage to check where you stand before leaning on it.

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

A senior engineer who interviewed 50+ AI-native developers lists the 8 common mistakes holding most people back.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 350K+ engineers and 150K+ followers on socials. Get in touch.

We put together a few guides on coding agents, agentic engineering, and leadership frameworks to help you level up in your career. Browse all our guides.

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

Until next time — The Code team