Cat Stefanovici

@bcherny is there a way to get commands and subagents working in cowork the way they do in Claude code? I opened a folder that had .claude commands and tried to invoke the command in cowork, crashed the app 😩😬

@csti Said another way…what the hell is everyone at Apple doing?!

our phones are still pretty dumb. here’s an example:

The OS should be applying AI to understand the purpose of each app (maybe apps can also have a manifest file specifically for operating system AI) and give you OS level notifications based on its understanding of the context of what your apps are doing. 💯

liquid glass is ... "interesting", but it’s not the evolution we need.

Key for large repositories.

@leenyburger Worth checking out @relume_io. They let you export to figma, with the export including a brand book.

"designing" (I use the term loosely) a big kid webpage for SaaS Marketing Gym. I've tried loveable, replit, and bolt and so far blah and bleh. Are there any more design focused AI code builders? Or "build me a landing page that doesn't literally look like every other landing page in existence"?

@laracasts G and H keep things consistent.

What's your favorite style for resolving a service from Laravel's container? 😀 // NJ

@NickADobos Magic squares are kind of captchas against humans en.wikipedia.org/wiki/Magic_squ…

Perhaps a reverse captcha of some sort? A puzzle humans can’t solve easily but ai can?

How can I build a website that ONLY operator can use? Any ideas? No humans allowed.

@svpino Cursor is next level. The composer + agentic mode coupled with an experienced engineer giving it iterative guidance is a productivity boon

Cursor is kicking Devin's butt and is not even close. The feedback on Devin keeps rolling in, and it's not good. So much for "the first AI Engineer." (And at $500/mo, you bet we'll hear more from companies and less from influencers who only care about hyping AI.) The Answer AI team tried Devin on 20 tasks, and it was a colossal failure. 14 tasks were unsuccessful, three were successful, and three had inconclusive results. And they couldn't find any pattern: Devin failed across the board despite the task's difficulty. I tested Devin 8 months ago. Devin loved to write code back then, even when it wasn't necessary. It was very verbose then and looks like it's still very verbose today. The Answer AI team documented every test they did at length. You can find the article below. Here are a couple of quotes from the team reflecting their frustration with the tool: "Tasks it can do are those that are so small and well-defined that I may as well do them myself, faster, my way. Larger tasks where I might see time savings I think it will likely fail at. So no real niche where I’ll want to use it." And also: "I had initial excitement at how close it was because I felt I could tweak a few things. And then slowly got frustrated as I had to change more and more to end up at the point where I would have been better of starting from scratch and going step by step." This is the third time I have read or heard feedback from a team that tried to integrate Devin but failed. I haven't heard from anyone who is using Devin successfully. Here is the kicker: The Answer AI team found that working iteratively with Cursor is much better and produces good results in tasks where Devin fails. It's early 2025, and AI + Humans is still the winning formula.

The only methodology that makes sense for how this was implemented: 1. an app clip (a lightweight version of an app that does not need to be installed) is triggered through an iMessage link 2. the app clip's sole action is to launch a live activity and disappear 3. the live activity remains visible on the lock screen and dynamic island until dismissed by the user According to Apple's docs: "Your App Clip can’t typically include app extensions, but starting with iOS 16 you can include a widget extension that offers Live Activities and displays live event data without requiring people to install your full app." 🤯🤯🤯

Frantically googling how @nikitabier got iOS live actions to display for @explodeapp message recipients that haven’t downloaded the app… 🤯 Anyone?

We start with some data from @_jellyfish_co who published the “2024 State of Engineering Management Report”, with survey responses from over 600+ professionals.

It’s 2025, are your software engineering practices ready for a leap? At Hyperformation, we enable you to grow faster than the rate of your own experience. This January, we’re diving in on trends in engineering practices: when and why to implement different tactics on your team

Most fun Captcha I've ever encountered. WebAssembly-based Doom port. UI built by @rauchg using V0. Tip: hold alt + (left/right) to strafe

@tibo_maker Launching peer groups for software engineering managers on @hyperformation

So what will you do differently in 2025?

@svpino You can set up stubs instead of using mocks. Using dependency injection to make the tested classes have the stub as a dependency means you get to have unit tests that are highly targeted and FAST.

I have several unit tests that take too long to finish (about 20 seconds each.) I don't know how to make them faster without mocking the heck out of my code. I don't like doing this. Mocks are a nightmare down the line. But these tests are slow, so I don't run them as frequently as I should. It's a shitty situation to be in. I still need the tests, but they are like rotten potatoes and make testing the code harder. For now, I'm just tagging the slow tests to run them separately from the rest. It's not perfect, but at least I get to iterate faster.