Generative AI replaced classifiers with a prompt. This week, the classifier came back with a vengeance in Jev from TypeSafe AI. Instead of generating text token by token, it scores decisions in a single pass. www.youtube.com/watch?v=G2Kt...

Jev AI Model Explained

Why the AI classifier came back: System 1 decisions and RLCD calibration with Jev from TypeSafe AI.

youtube.com

You're burning tokens every time your AI agent guesses CLI commands instead of checking syntax. Do essential skills keep it on track? I ran an ablation study on 12 Cloud use cases to find out: medium.com/google-cloud...

Do agent skills help? I ran 72 trials to find out

An evaluation of the Google Cloud Developer plugin with NVIDIA SkillEvaluator and Harbor

medium.com

The starter pack for Google Cloud is here. Install the new agent plugin to your favorite agent: • Real-time docs via MCP • 100+ skills on demand • Auth & onboarding guardrails Watch the breakdown: youtu.be/_qVCARkqR30

Trying Antigravity isn't an all-or-nothing choice with the new IDE extensions. Bring Gemini's multimodal reasoning into: • VS Code • JetBrains IDEs • Xcode • Zed • Visual Studio Extensions are a place to stay, or a place to start. antigravity.google/blog/antigra...

I love vibe coding as much as anyone, but what's the one thing you would NEVER let your agent hand-roll from scratch? 5 things I always use a service or library for: 1. Authentication 2. Payments 3. Dates & time zones 4. Cryptography 5. Job management What's missing?

We just shipped Gemini 3.8 Flash! It's our third Flash model in six weeks. It really shines at long-horizon coding and autonomous agent tasks. To showcase the improved reasoning and tool-calling, check out this game written with a single looping prompt. Prompt and launch post are in the thread 👇

Rolling your own auth is tech debt waiting to happen. Use @firebase.com Authentication as an identity component and integrate it with your existing stack. Free for the first 50,000 MAUs for email, social, and anonymous sign-in. Read more in my latest post: firebase.blog/posts/2026/0...

Authentication made easy: Building a secure e-commerce shopping cart with Firebase

News, tutorials, and updates from the Firebase team.

firebase.blog

Prompting AI for a "7-day Italy itinerary" is a recipe for tourist traps. Orchestrating an integrated suite of AI tools is how you build an unforgettable trip. When my wife and I planned our anniversary trip, we chained together 4 tools in this playbook.

How I Planned an Italy Trip with an Agentic Workflow

Using deep research, a browser agent, and calendar integration for trip logistics

substack.com

I was about to build a multi-agent system to assemble my monthly newsletter. I wrote a skill file instead. A year after AI agents hit the enterprise, a lot of developers (myself included) fall into the over-engineering trap for personal workflows. medium.com/google-cloud...

Why I write skills instead of agents for knowledge work

Are there tasks you find yourself doing repeatedly, wondering if there is a better way? For me, one of those is writing a bi-weekly…

medium.com

You don't need a math degree to vibe code with AI, but knowing just enough will save your app from freezing up or giving you unexpected results. I wrote about 4 math concepts every vibe coder needs: probability, tokens, vectors, & Big-O: www.linkedin.com/pulse/four-m...

Four math concepts to improve your vibe coding

The power of vibe coding is undeniable. Describe what you want in Google AI Studio, Antigravity, or your favorite agent, and you'll get it! You don't need to know the math behind why things work the w...

linkedin.com

Matt Pocock's Skills 1.1 just dropped: 🧭 Wayfinder: progressive discovery for multi-session work 🔬 Background research agents 📋 Consolidated planning pipeline 🔍 12 Fowler smells in code review 🎯 Grilling won't answer its own questions ✂️ TDD trusts the agent npx skills@latest add mattpocock/skills

Your dependency bots are doing their job. Something needs to do the next step. So I built an agent that does: merge if green, fix if red, flag if stuck. medium.com/google-cloud...

The future of autonomous software maintenance: a dependency update agent

Most AI tools focus on writing new code, but up to 80% of the cost of software comes from maintenance. What if an intelligent agent could…

medium.com

How can corporations support open source maintainers? Beyond direct sponsorship, what creative funding models actually help you sustain your work? Let's discuss. goo.gle/improving-op...

goo.gle

How can corporations better support open source maintainers? Explore feedback from open source communities on creative funding.

goo.gle

2 major model releases today! You can generate cost-efficient images in 4 seconds with Nano Banana 2 Lite (GA). And with Omni Flash, you can generate video with real-world knowledge and perform conversational video editing (Preview). And yes, the infographic was created with Nano Banana 2 Lite!

Most developers don't think of AI agent skills as dependencies. But they behave just like them. An agent skill is a package of instructions and scripts that help an agent accomplish tasks. Just like code libraries, these instruction sets drift as the underlying APIs and systems evolve.

Many developers treat testing as an afterthought when using AI coding agents. "Build this feature. Now write unit tests." The result? Green test suites that give a false sense of security. Here are 5 essential tests your agent should create instead. www.youtube.com/shorts/3mm_8...

Essential Testing Strategies for AI Agents

I share why most developers fall into the coverage trap when prompting AI coding agents and how I treat test suites as architectural guardrails. In this walkthrough, I break down five essential…

youtube.com

Infrastructure setup is the ultimate "stop energy" for new ideas. How many times have you abandoned a side project because configuring billing accounts, APIs, and security poliicies stalled your momentum? The Starter Tier on Google Cloud demolishes this friction 👇

I deployed an app from localhost → dev → production using an AI agent. No bash scripts. No gcloud commands. No YAML. Watch the full workflow 👇 youtu.be/cFSxmXQnU5Q

Antigravity: The AI Agent that Runs DevOps for Me

YouTube video by Cloud with Karl

youtu.be

"Knowledge-as-wiki" is an emerging pattern Karpathy made popular with LLM Wiki. Google has formalized this with the Open Knowledge Format. OKF gives agents a common language to understand your APIs, data schemas, and more. My latest article breaks it down. www.linkedin.com/pulse/your-a...

Your AI agent reads files. It should use OKF to read systems.

Today's AI coding agents don't hallucinate the way they used to; we're generally past the "inventing fake APIs" stage. The failure mode now is subtler: they over-index on part of the codebase and miss...

linkedin.com

Huh… @kweinmeister.bsky.social, people seem to like Gemma-4. Lon gave the 26B a try. youtu.be/uOHI47l215U

Local AI is Finally Usable! Real-World Workflows on a 5-Year-Old Mac with Google Gemma

YouTube video by Lon.TV

youtu.be

The Google Antigravity SDK enables you to embed Antigravity agent capabilities directly into your codebase. My latest article walks through how to get started, then explains each architectural concept in the SDK. www.linkedin.com/pulse/google...

Google Antigravity SDK: The developer guide

The Google Antigravity SDK is a Python framework for building and running autonomous agents. It decouples your agent’s logic from where it runs, letting you focus on what the agent does while the SDK ...

linkedin.com

I will show you how to prevent your AI coding agent from building the wrong thing. Use the /grill-me skill so it knows what you really want. A few extra questions upfront saves lots of refactoring and frustration later.

The happy medium is here in Gemma 4 12B: powerful capabilities at a size that fits well on your laptop. Multimodal, Apache 2 license, reasoning, and much more.

Pythonistas: which type checker do you like best? ty is my go-to, and I'm exploring pyrefly.

Can a hand-drawn napkin sketch become a working app in under 3 minutes? I uploaded an AI Audio Engineer App sketch to Google AI Studio and let Gemini 3.5 Flash write all the code. #NapkinChallenge www.youtube.com/shorts/jgB8x...

Can AI code this sketch? AI Audio Engineer App

Can you turn a scribble on a paper napkin into a fully functioning, beautiful web application in under 3 minutes? In this video, I take on the Napkin Challenge! With any rehearsing, I uploaded a…

youtube.com

I continue to be amazed every time I use Nano Banana 🍌. Today marks the General Availability release of Nano Banana 2 and Nano Banana Pro on Gemini Enterprise Agent Platform! 🎉 cloud.google.com/blog/product...

Nano Banana 2 and Nano Banana Pro available for everyone | Google Cloud Blog

Nano Banana 2 and Nano Banana Pro are generally available today via Gemini Enterprise Agent Platform. These models empower you to integrate high-quality image generation and editing capabilities…

cloud.google.com

Medium doesn't support markdown tables. So I built an MCP server that converts them to styled PNG images right from Antigravity. Wrote up how the rendering works and how to build your own. medium.com/google-cloud...

When evaluating a long-running AI agent, the top 3 metrics to watch are: ✅ Success rate: was the goal achieved? ⏳ Duration: how long & how many steps? 💸 Cost: how many tokens were consumed? What would you add to the list? www.youtube.com/shorts/rebAE...

AI Agent Metrics that Matter

At 95% per-step reliability, a 20-step agent succeeds less than 36% of the time. I cover the three metrics — success rate, token usage, and duration — that tell you where your agent is actually…

youtube.com

Gemini Spark - all new AI agent. It’s going to help me stay organized and even create slide decks. #GoogleIO

Can’t wait to try out the Gemini Omni model. Stay tuned for a video on how to edit video with it! #GoogleIO