André Bandarra

A browser agent meets a website's own agent. Tool call, handoff, or real conversation? 🤔 I built a demo - WebMCP + A2A. Turns out the protocol is the easy part; the real question is who thinks, who owns the task, and who pays 👇 kulikowski.me/blog/agents-ta… #AgenticWeb #WebMCP #A2A

Meet LiteRT.js: @Google’s new Edge AI runtime for the web! ✨ We've made it easier to convert from PyTorch to #WebAI using the common LiteRT model format. Powered by #WebGPU, WebAssembly, & #WebNN! Time to upgrade from TensorFlow.js and build the future. Read more ➡️ goo.gle/4eUP97n

Three homes of the Agentic Web by @i_am_kaspa kulikowski.me/blog/where-doe…

Join @Lojanda and me on this deep dive into the intelligence system and Android MCP using the AppFunctions API: youtube.com/watch?v=qtQMH8… #AndroidDev

AI agents are changing how users interact with the web. Are you ready? 🤖🌐 To join in the Agentic Web you need "agent-readiness." Focus on Discovery, Understanding, and Usage to make your site accessible for AI. Read more at bandarra.me/posts/the-agen… #AgenticWeb #WebMCP #AI

@SeraAndroid Rule of thumb: use the smallest model that is just good enough to solve your problem.

I recently joined an AMA session about AI in Johannesburg. One question that caught the audience's attention was the challenge of feeling lost when dealing with token cost. My recommendation? Don't use the latest frontier model to solve all problems. Let it be your planner, your orchestrator. Let it breakdown the ambitious goal into tangible steps. Hand those steps and plans to an executor. This model drives the implementation and doesn't require the same amount of intelligence. Finally, let a third model verify and judge the output of the executor. This model doesn't need the same capabilities as the orchestrator and the executor. It has access to the plan, implementation, and reasoning of the other two models and can work with that to assess if you've managed to hit your goal.

The best way to learn and shape opinions is to build things. So, I started a personal blog to share my thoughts, demos, deep dives, and occasional hot takes 🌶️😅 First up: Unpacking the Agentic Web, less technical, more general thoughts: kulikowski.me/blog/what-is-t… If you’re building or experimenting in this space, I’d love to connect and chat! 💭 #AI #WebDev #AgenticWeb

@sarah_edo Yeah, and I think "pair prompting" as a way to help junior engineers gain experience will become a thing.

The way we train juniors will focus more and more on system design even at beginning stages

Some vague predictions: We pick our models today but as tokens and context management become more important, there will be more abstractions to automate and create efficiencies

@harshil1712 CC @i_am_kaspa

Diving into WebMCP at #GoogleIOConnect led by @andreban and Kasper! #BuildWithGemini

@cramforce BFR9000

What should we name Vercel regions with >1 million vCPUs?

📢 Breaking news: @googlechrome folk just published a guest post on @huggingface 🤗 blog around the newly proposed Cross-Origin Storage API and how it's being used in #TransformersJS! This allows huge #WebAI models to be shared across domains! Read 👉 lnkd.in/e_rgfwK6

@dalmaer This is more about the parent's ability to set limits and boundaries for kids than it is about remote working.

This is the dumbest thing I have seen. Remote work doesn’t have to mean “you have to work in a room with your kids running around you”. 🤦‍♂️

Whether your are building an AI Agent that uses WebMCP tools or exposing tools on your site to AI Agents, there's security guidance to take in consideration! #WebMCP #AgenticWeb

New polished docs and live, on-device demos for the Web AI SDK 👨‍🎨 Check it out: web-ai-sdk.dev

The WebMCP Origin Trial in Chrome is here! Try it out today! We're looking forward to your feedback! #WebMCP #AgenticWeb

@TheKryptoWiz Agreed! Ensuring new features and APIs on the web work well with frameworks and are well documented is crucial to their success.

@andreban Framework-native tool exposure is the part that makes agents practical. Once WebMCP sits in the docs, teams can ship useful surfaces without inventing glue code.

Are you building with Angular and wondering how use WebMCP to expose tools to AI Agents? The Angular team has got you covered and published detailed docs on how to use WebMCP with the framework! Read all about it at: angular.dev/ai/webmcp #WebMCP #AgenticWeb

Yes, Doom runs on AI agents using WebMCP. Video from my recent session during @ainativedev. "Play Doom as a maniac for 1 minute" #ainativedevcon

@firt @ainativedev See you tomorrow! Looking forward to watching the talk and chatting about WebMCP!

Nice weather in London these days. I'll be speaking tomorrow at the @ainativedev conference on WebMCP. 😎