Brian Krabach

New in Chrome DevTools: Annotate performance findings with Gemini! We just shipped a brand new feature in DevTools. You can now ask Gemini to explain performance traces. Specifically you can generate annotations about events in the performance trace. This is a total game-changer. Double-click an event in the Main track, then click Generate label next to the input field. Gemini can suggest a label based on the stack trace and context. Try it out and let us know what you think! #ai #programming #softwareengineering

1,233

54 Comments

Seeing ASP.NET WebForms revisionist historians, kinda like folks saying "actually the Star Wars prequels were good!" No, they're not. And no, Blazor isn't WebForms, nor does it vindicate WebForms' design. WebForms were an abstraction to pretend the web doesn't exist in favor a completely different, fundamentally incompatible model. Yeesh.

129

44 Comments

A very deep problem in music-related software has been really highlighted in the AI era. That problem is what I would consider the “Data Model Problem.” Take a MIDI file, for example. It has stamped events in time, like when the note starts or stops. It’s pretty good at representing a sequence of notes recorded by the user on their keyboard and playing them back later. But if we want to build some high-level program that helps the user do something useful in the composition space, MIDI is actually really wrong. For one, it can contain any sequence of any notes in time, including unmusical notes, mistakes, etc. But it also contains nothing about the intention behind the recording. For example, say I generated some MIDI clip based on a chord progression — that chord progression is gone from the resulting sequence. In most sequencers (such as our Beat Scholar, EON-Arp, etc. plugins), this problem is handled by creating a different data model to represent the music. For example, one might decide that the sequencer has a grid (like 16th notes) to limit what the user can input. That makes it clear not only when the note is meant to happen in time but also how the user is thinking about that note placement, which becomes important if you want to apply some time-based effects like swing but maintain the original relationship. Now that AI can be somewhat involved in that space, I think that problem is more important than ever. We need to think of data models that, when observed or read by the user, contain meaningful representations that aren’t just big data blobs. The users of such software need to be able to look at that output and make forward progress with it, just like they would when creating or editing the data manually. It needs to be something that closely resembles a “process” and “intent,” giving the user hands-on understanding so they can use it for personal and artistic growth, even when incorporating external content like presets or AI-generated content. What are your thoughts about this problem? Have you found interesting platforms where it was tackled in a way you thought was a good step in the right direction?

40

28 Comments

Never "act very rudely" with accounts you sign into - particularly if you have a visible session history. Why? Because even apps that don't currently support memory (like Claude, as of this writing) may add memory across your previous chat history later, and the model will see you acting like a jerk, and that could flavor future outputs. You wouldn't be rude to your coworkers, right? But just like you should treat humans and coworkers nicely, it can't hurt to do the same for AI. Even if they're going to stay "a computer that talks" for a while 😁

47

26 Comments

This is an interesting analysis over at RedMonk of how Wasm is working through the growing pains in and beyond the browser. And Kate Holterhoff, Ph.D. also introduces the features of WebAssembly 3.0. I feel obligated to mention that the Component Model is NOT part of Wasm 3 because it is standardized as part of WASI, which will see its p3 release very soon. The theory in separating thus was that not every Wasm runtime needed to implement the component model, which has more of the system interface feel to it. Most Wasm 3 features are already supported by Spin and Wasmtime, with the rest coming as soon as the PRs get completed and merged. https://lnkd.in/g2YWCTrb

32

2 Comments

🎶 I’m excited to return to speak at #Nodes2025, the biggest graph community gathering dedicated to applications, data intelligence, knowledge graphs, and AI. This year, I’m continuing the music graph story that began in our book Neo4j: The Definitive Guide, now powered by Spring AI. From “influenced by Pink Floyd” to “worked with electronic producers”—we’ll explore how graphs + AI unlock powerful music discovery. Why Spring AI? Because AI today isn’t just about models—it’s about building robust, reliable, secure and composable systems. And the JVM ecosystem has long mastered reliability, observability, and integration at scale. AI needs that now more than ever. When it’s time to deliver real value in production, I’ll argue the JVM is having a quiet resurgence with frameworks like Spring AI and Embabel. And that’s worth talking about. It's an online event, see you there! https://lnkd.in/dJchw2n4 #Neo4j #GraphAI #SpringAI #Nodes2025

77

6 Comments

Most online services today run over HTTP, great for requests and responses, but not so great for real-time audio/video. WebRTC fills that gap, yet for years developers had to wrestle with clunky Chromium wrappers or GStreamer. That’s finally changing. Native WebRTC libraries are now available in languages like Rust and Kotlin. The Rust webrtc crate is far simpler than past solutions, and Ktor 3.3.0 brings WebRTC support to Kotlin. This makes building real-time apps (far beyond just VoIP) much easier in two of today’s most modern languages. Curious about what you can do with WebRTC? I’ve been working in this space for years and have seen some exciting possibilities. 🔗 https://lnkd.in/dQ6jSygW 🔗 https://webrtc.rs/ #webrtc #rust #kotlin. #ktor

53

8 Comments