The smell of vibe coding (News)

0:28What up nerds? tear-inducing videos. Go ahead, click to feel something. All dried out? Okay, let's get into the news. The smell of vibe coding. Alex Kondov knows when you've been vibe coding, he can smell it. Quote, no one would write an HTTP fetching implementation covering all edge cases when we have a data fetching library in the project that already does that. No one would implement a bunch of utility functions that we already have in a different module. No one would change a global configuration when there's a mechanism to do it on a module level. No one would write a class when we're using a functional approach everywhere.

1:09End quote. See, he's on to you. Ultimately, Alex doesn't care about how the code got into your IDE. He just wants you to care. To care about quality, to care about consistency, to care about the long-term effects of your work. And if you care, quote, don't leave a code base's maintainability to the weights of a model. A glam AI coding agent for your terminal. Our friends at Charm are at it again. This time they've built a Go-based AI coding agent as a TUI. It's called Crush and it's multi-model, so you can choose from a wide range of LLMs or edge your own. It's flexible, so you can switch LLMs mid-session while preserving context.

1:47It's session-based, so you can maintain multiple work sessions and contexts per project, and it's extensible, so you can add capabilities via MCPs. Disassembling the T-app hack. There's been a lot of speculation and a whole lot of jokes that T, an app used by women to dish about men on dating apps, was vibe-coded because of just how epically and easily its data got leaked. Jen Kamerath disassembled the Android app's binary and concludes, not so. Quote, my assumptions after this initial forensic analysis is that this app was built by a single, inexperienced developer or by a team dictated by a single, inexperienced developer.

2:31The app was likely not vibe-coded as none of the models of the past months would have made such obvious mistakes. End quote. Ouch. In other words, vibe-coding would have produced better results. quote this hack should have never happened the t app is an ai slop it's gross negligence from a likely single developer with very little experience that should not have been allowed to publish such an application without supervision the app didn't get hacked it willingly published sensitive personally identifiable information to the world it's now time for sponsored news observability for your GitHub actions.

3:10If you ever stared at a failing GitHub action, wondering what on earth just happened, you are not alone. CI logs are often a black box and not the fun kind. That's why Depot just launched GitHub job details, a new observability layer for your CI CD pipeline. It lets you zoom in on what each GitHub actions job is actually doing real build times, dependency fetches, computer performance, and more. No more guessing, no more click fests through the raw logs. Your team gets better visibility into bottlenecks, misconfigured jobs, or flaky performance without rerunning things a dozen times. It's observability built for CI, not just another dashboard.

3:49Check it out at depot.dev and thank you to our friends at Depot for sponsoring Changelog News. Price per token. Is your LLM spending budget on the rise? If so, Alex Ellman has just the website for you. Price per token provides up-to-date pricing info for major LLM APIs, including OpenAI, Anthropic, Google, and more. It pulls data from OpenRouter.ai and even lets you estimate the cost of executing the same prompt across different models. Plus, they just added image generation comparisons. In the future, I assume every AI tool will have a smart routing layer up front that dissects our prompts, determines the sweet spot provider of just enough quality at the cheapest possible price, and dynamically routes it on our behalf.

4:36In the meantime, tools like this one from Alex could save you a bundle. If you're remote, ramble. Steph Ango, of Obsidian fame, makes a suggestion for remote teams of 2 to 10 people like his. Quote, create a personal ramblings channel for each teammate in your team's chat app of choice. Ramblings channels let everyone share what's on their mind without cluttering group channels. Think of them as personal journals or microblogs inside your team's chat app, a lightweight way to add ambient social cohesion. End quote. These channels should be at the bottom of the list, muted by default, with no expectation that anybody ever reads them.

5:17The Obsidian team has found them surprisingly sticky. Quote, because they are so free and loose, some of our best ideas emerge from ramblings. They are often the source of feature ideas, small prototypes, and creative solutions to longstanding problems. That is the news for now, but go and subscribe to the Changelog newsletter for the full scoop of links worth clicking on, such as most language migrations are hype-driven, modern Node.js patterns, And typed languages are a better fit for vibe coding. Get in on that newsletter at changelog.news. In case you missed it, last week on the pod, we hosted Greg Osuri on solving the AI energy crisis.

6:00That episode has already generated a lot of discussion in Zulip. And on Changelog and Friends, Adam and I sat down with the 2025 Stack Overflow developer survey results to glean what we could glean. give those a listen scroll up in your feed or down i guess depending on your sort order we have some great ones coming up this week too both of our denver live shows are shipping our interview with norah jones comes out on wednesday and on friday kaizen 20 with gerhard lazou have yourself a great week like subscribe and five-star review us if you dig the show and i'll talk to you again real soon