Master.dev (Formerly Frontend Masters)

@rahulkumarpahwa please email support@master.dev so the team can look into your account/issue. thanks!

Hi @MasterDotDev team! I hope you're doing well. Some of the videos in the new courses aren't playing when I try to watch them. Could you please check when you have a chance? Thank you!

How to give an AI agent live awareness of canvas state, in one function: "We just need a function where we can extract elements that we sent up on the user message from the data parts type, so we can inject it into our stream agent. Extract canvas state, it's going to take in the UI messages. The last message in the array is always the user message, so we can get messages at negative one. Find any part where part.type is data-canvas-state, return those elements. I'm going to put a square, an arrow, a circle. Tell me what you see on the canvas. I see three elements on the canvas, a rectangle, an arrow, and an ellipsis. Perfect."

Weaker models with a great harness beat stronger models with a bad one. The model sets the floor. The harness sets the ceiling. "The models are all pretty good now. I think whoever builds the best harness will win." ~Scott Moss

What is an agent harness? Claude Code is a harness. Cursor is a harness. Codex CLI is a harness. The model does the thinking. The harness does everything else: memory, tool calling, durable execution, handing work to other agents. An agent = an LLM + a harness

CDNs explained in one clip: why your website loads faster depending on where you are in the world. "Instead of having one bucket, one place, what if we put very small infrastructure all over the place? The first user will pay the cost of going all the way to the initial bucket. Let's say you're on the west coast of Australia and somebody on the east coast already went to go hit that page, you don't have to go all the way to DC, you just need to go to the East Coast for that regional cache. To no one's surprise, it worked better the closer you were to Chicago, and progressively worse the further you got away. Somebody comes in, they go to the nearest one, do we have the cache, if so serve it, if not go grab it from US East 1."

RSVP for our free 5-day AI Building Bootcamp with @brmiddle and @holtbt. Zero coding background needed, just bring an idea. You'll walk away with a real deployed app and the fundamentals to keep building. Starts July 20th: master.dev/workshops/vibe…

@alire8za Ohhh ya @scotups is definitely cooking up some more awesome AI courses!

@MasterDotDev Great course by Scott Moss. Highly recommended. I hope he is invited back with more AI engineering courses

A naive AI agent with a "send reply" tool will just send your drafts. All of them. Without asking. Here's the actual failure mode: "Imagine you had some business emails and your LLM was like, cool, I'll draft these for you. But because it has a send reply tool, it just decided the objective was to reply to all these emails, and I can't be done until I sent them, so I'm just going to go ahead and send these drafts off. You might ask, what if I just put a description that says don't use this tool unless you ask the user first? That's not how it works. If you're doing guardrails and safety stuff in text, you're gonna fail." Real constraints have to be structural, not a sentence in a prompt.

OMG, this is such a good framing: you can chose if you want to use LLMs to generate more code or better code

Rich Harris on what the best engineers actually do with AI: "Writing code was never the bottleneck. Understanding what you're building is the bottleneck." Using AI to think through the problem beats using it to just generate code. Full interview: youtube.com/watch?v=SKjXE_…

Claude Code has no memory. It rebuilds context from zero on every call. Here's the actual mechanism: "The model is stateless. Every time we call the model, it starts from zero. The moment you press enter, Claude Code assembles all of this data... this tells the model who it is, what tone it should have, the coding convention, security rules... The model doesn't read JSON, it's up to the API to internally turn this into tokens the model sees. The user wants to add a test in utils.cs. Nowhere in our assembled prompt have we added this file yet. In our tool schema we have a read tool call, so it understands: if I don't have access to something, use the read tool call. When it's successful, it reassembles the entire prompt again." No memory. Just context, rebuilt every turn.

Free 5-day Building with AI Bootcamp, July 20th-24th. Build the skills of the future and get ahead. In just one week, you'll go from zero experience to building real projects using natural language and the latest AI tools, guided by fundamentals that set you up for long-term success in a fast-moving industry. No coding experience needed! 👉 master.dev/workshops/vibe…

TanStack Router's dev tools give you some wild type safety "What if I put incorrect stuff in here? TypeScript starts screaming at me immediately. It knows exactly what it wants, down to the trailing slash. I hit Command Save. TanStack router said, nope, I saw your change, and I reject it. That is because of the dev tools that are running. Let's create a new route, foo.tsx. Bam. It scaffolded everything for you, the simplest possible route, with all your options and the correct path. If you rename it, it renames the route. The dev tools are fantastic."

AI coding tools are useful. But Evgenii Ray thinks they're still missing the point. Here's what closer integration with your dev environment could actually look like. Full interview: youtu.be/Hwb4TJfjfb0?si…

Giving Claude Code access to your codebase doesn't mean giving it access to everything. Deny rules let you draw hard lines. "But now I also want to add a deny rule and I just want to see what happens. So if I want to deny, maybe edit package.json. And now I'm going to still ask Claude to edit it for me." "So I'm just going to actually change this model to haiku because again, this is just a very simple task. So I'm going to like update the version in package.json to one. Now hopefully my permissions were set up correctly and it will be like, can't do that." "Yeah, so it says like, oh, I can't do that. So even if this was a tool call coming from the model, it won't run it. So this is like a great way to add more control over what Claude can and cannot do in your environment."

@AdHominemIzBad @kagigz @OpenAIDevs Could you please email support@master.dev with more details? Our team is has not been able to replicate this error.

@AdHominemIzBad @kagigz @OpenAIDevs Our native mobile app? We’ll take a look and fix! You can also email support@master.dev to get real time updates on the fix

On Aug 11, I'm hosting an @OpenAIDevs workshop with our friends at @masterdotdev covering how to build advanced frontend workflows with Codex. We'll go beyond basic prompts to explore how you can use Codex with the right setup for frontend dev. Sign up link below 👇

Rich Harris on AI in his dev workflow: "Framework design is hopefully exploring new space that hasn't yet been seen by AI before." Also his underrated take: AI is better at reviewing code than writing it. Full interview: youtube.com/watch?v=SKjXE_…

Vibe coding a demo is easy. Making it dependable is the whole job. @‌scotups breaks down what AI engineering actually is. Context engineering. Tool design. Evals. Production reliability. Learn more in our AI Engineering Fundamentals course.