Recently posted how I used Google's #Antigravity IDE with #Gemini to make my own incarnation of Google Earth in a couple of hours. While amazing, there were a few UX issues I wanted to fix. Well I fixed them in a few more hours, added brand new features, too😯 - <70KB of front end code! #WebDev #JS

Casually training Auto Encoder on the client side via #WebGPU using TensorFlow.js to try some idea out. Forgot how much I loved using TFJS - client side AI since 2017 :-) What you see here is the encode / decoder training in real-time. I can then chop out decoder & gen those imgs. Fun! #WebAI

Should we make these? Having weekend fun with Gemini / Nano Banana. Would be great event swag, wdyt? The only device you need to make dreams come true with #Gemini + #Antigravity. I should probably add a tab key too😅 What would you add to make it prod ready for agentic coding?✨

New technical disclosure published via work at @Google with #TDCommons: "Orchestrator-Worker LLM Architecture for Mitigating Prompt Injection via Separated Privileges". How to reduce number of successful prompt injection attacks for some agentic flows: www.tdcommons.org/dpubs_series...

The #WebAI Monthly #32 (May edition) is now live: 💰 Looking for a bot in finance? Meet Wilson. 🎞️ Adobe Premiere clones are here, and dive deeper into #WebMCP - the future of the agentic internet 🤖 www.linkedin.com/feed/update/...

Web AI Monthly #32: 📊 Looking for a bot in finance? Meet Wilson. 🎞️ Premiere clones are here, and dive into WebMCP - the future agentic internet 🤖 | Jason Mayes

Web AI Monthly #32 (May edition) is now live: 💰 Looking for a bot in finance? Meet Wilson. 🎞️ Adobe Premiere clones are here, and dive deeper into #WebMCP - the future of the agentic internet 🤖 📖 Rea...

linkedin.com

What does an agentic internet look like? 🤔 At last year's conference @webai.bsky.social walked us through how to create and run client-side AI agents in JS, powered by smaller LLMs, to perform useful work faster on existing websites using Web AI. Take a look! youtu.be/4J9RaiomJFo

✨Recently took part in panel discussion with @AICollectiveCo in #SanFrancisco, and some deeper 1-1 interviews, with questions submitted by their community. Join us as we dive into the future of the agentic web, #WebAI, and of course #WebMCP!➡️ www.youtube.com/watch?v=jZyN...

Collective Intelligence | Shaping The Future Of AI with Google's Web AI Lead, Jason Mayes

YouTube video by The AI Collective

youtube.com

After using the @EvenRealities #G2 AR smart #glasses & #R1 ring for > 1 month, it's review time! How do they compare to the G1s? Are they worthy of upgrading? Join me as I delve into the features & performance of this futuristic eyewear: www.youtube.com/watch?v=nT7L... #Tech #Review

Even Realities G2 Smart Glasses and R1 ring review - 1 month in.

YouTube video by Jason Mayes

youtube.com

➡️ I created a custom skill to give #Gemma 4 ability to extract information from any website. 👩‍💻You can then say stuff like: "Give me TLDR of this guy's experience at jasonmayes.com using the first 3 paragraphs of text or headings" 🔗 Link (vote for it!): github.com/google-ai-ed...

✨ Exciting news from Google - #Gemma 4 is here! 🪟 Up to 256K context window for larger variants (26B / 31B) ⚒️ Built for action & agents - tool use & multistep planning. 💻 Designed for edge AI inference (E2B / E4B) with near zero latency and have 128K context window! deepmind.google/models/gemma/

#WebAI Lib for semantic vector search, search millions of vectors in milliseconds, and visualize the tokens or embeddings too! Runs entirely client side in web browser (custom Vector DB written on top of IndexDB) Supports EmbeddingGemma & all-Mini-L6-v2!⭐️https://github.com/jasonmayes/VectorSearch.js

❓I can ask questions for what it sees to save time. This image shows it on my #LinkedIn profile to figure out best way to speak to that person - even got my MBTI personality correct!➡️ Everything runs entirely locally, so privacy when sharing pages that you may be signed into!

🤯I finally figured out how to get our client side #WebAI multimodal #Gemma 3N model to play nice in a #Chrome extension ✨ Now I can have a smart companion that can work with me in any tab I choose to share to it 🤖 I can even save examples that work well as "recipes" to reuse.

Introducing VectorSearch.js (star it & I'll continue to work on it)! A JS lib that allows you to perform semantic search over millions of vectors in ms, & can visualize embeddings too! Runs #Google's popular #EmbeddingGemma model client side! github.com/jasonmayes/V... #TransformersJS #LiteRT #TFJS

Everything to make this self generating code execution demo work is running client side in a web browser using #WebContainers + #WebAI. Total privacy, sandboxed code execution (Node.js via Wasm), zero cost. Using off the shelf 2B Gemma model (not tuned) 69pjnq.csb.app

🚨 Demo Alert: Today I met with a really interesting company, bolt.new, who's involved in #WebContainers.🫙What's that? Imagine you could run Node.js in JavaScript on the client side via #Wasm.✨Time to combine with #WebAI right? Yeah I thought so too... Try it: 69pjnq.csb.app

5 Defensive Publications for #Google. New milestone award unlocked to add to the collection ✨ If you're curious what they were on, check my LinkedIn profile under publications -from digital teleportation to physics based encryption - fun times - stay curious & invent the future!

My #WebAI Summit talk "Built-in AI in the wild: A Mastodon translation success story" is now live: youtu.be/-l2BUUdBDS8?.... I had so much fun giving this talk, and in it I teach you the German word for waving back at someone who wasn't waving at you: Fremdwinkzurückwinkungspeinlichkeit 🇩🇪 👋 😳.

Built-in AI in the wild: A Mastodon translation success story

YouTube video by Chrome for Developers

youtu.be

🤯 Just updated my depth demo for #LiteRTJS to create 3D point cloud of webcam feed in real-time entirely client side! ✨ As this is #WebAI everything is running entirely locally in browser via #WebGPU using #Qualcomm DepthAnythingV2 model. Also means you can try it yourself here: lnkd.in/gDutNjjC

Missed the #WebAI summit or need a recap? We've got you covered ▶️ Catch the high-level opening talk from Google's Web AI lead, @webai.bsky.social to understand the state of Web AI today → goo.gle/4rgxLOo

✨This year Google launched LiteRT.js - a client side Web AI runtime that can run AI models in browser! To learn ropes I found a depth model from #Qualcomm that runs in Chrome via #WebGPU! Greener pixels mean closer to the webcam, and bluer pixels are further away. ➡️Try it: codepen.io/jasonmayes/p...

Just used #Gemini3 to redesign my website that I've not had time to touch as so busy with #WebAI ;-) Is it perfect? No. Is it an inspiring start that I can fix up? Hell yeah! Thanks @developers.google.com. Demo: codepen.io/jasonmayes/f... #Portfolio #Redesign #Gemini3 #Gemini3Pro #Google #VibeCoded

Sound on! Just made a real-time MP3 visualizer in JS with #Antigravity! I decided to pivot to diving deeper with Antigravity (a brand new agentic development platform) by Google to work on something longer than 60 seconds. This was the result: www.linkedin.com/feed/update/...

#antigravity #gemini3pro #javascript #html #css #vibecode #google #gemini #ide #webdev #javascript #engineer #ide | Jason Mayes

Sound on! I just made a real-time MP3 visualizer in JS with #Antigravity! Live demo linked below. After trying raw #Gemini3Pro at our launch today in my prior posts, I decided to pivot to diving deepe...

linkedin.com

Wow. 🤯 Mind blown. Used Gemini 3 to create this retro flight game complete with collision detection and retro sounds that react to speed of plane. How long did it take? About 120 seconds. Vanilla JavaScript, fully functional on first try, I am impressed. Try: codepen.io/jasonmayes/p...

Pleased to say our #WebAI summit videos are starting to go up for 2025! Catch up now to get the latest scoop on client side AI in the browser. 7 talks up, 13 to come - so sub to playlist to get notified when they do www.youtube.com/watch?v=360P... #Agents #AI #WebMCP #JS #WebDev #Chrome #Google

Web AI Summit 2025: State of client side AI

YouTube video by Chrome for Developers

youtube.com

Grinding my gears: Folk creating new standards (#TOON) vs using existing correctly (#JSON). Going virial right now is not an apples to apples comparison. Can write VALID JSON in just 26 tokens (< TOONs 35). TOON is 34.6% worse here. www.linkedin.com/feed/update/... #MCP #LLM #Tokens

Whether you join in person or watch on YouTube later, you won't want to miss the #WebAI summit! We'll have deep dives on client-side ML, demos of new web tech, & networking opportunities with industry leaders. Grab your spot with code WEBAISOCIAL → goo.gle/WebAISummit2...

Ready to see latest in client-side #AI?🤯Join us in-person for the #WebAI Summit at #Google Sunnyvale on 17 Oct! Who: Speakers from Huggingface, Arm, Intel & more! Explore the latest in on-device Gen AI, real-time ML, & AI agents. Spots are limited! Use code WEBAISOCIAL here: goo.gle/WebAISummit2...

Jason Mayes (@webai.bsky.social) is coming to #CascadiaJS to show us how to create and run client-side AI agents in JS, powered by smaller LLMs, to perform useful work faster on existing websites using Web AI. ⚡ Get a glimpse of the future—an agentic internet. 🤖 More info 👇

Jason Mayes (@webai.bsky.social) is coming to #CascadiaJS to show us how to create and run client-side AI agents in JS, powered by smaller LLMs, to perform useful work faster on existing websites using Web AI. ⚡ Get a glimpse of the future—an agentic internet. 🤖 More info 👇