shakedown.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
A community for live music fans with roots in the jam scene. Shakedown Social is run by a team of volunteers (led by @clifff and @sethadam1) and funded by donations.

Administered by:

Server stats:

269
active users

#cursor

0 posts0 participants0 posts today

As a general rule, I always .gitignore files which can be regenerated; things like test run results. But now that I'm working with AI assistants that may be unwise. It's sometimes time consuming for them to fully install the package and run all the dev scripts just to get a file they need to have in their context in order to complete a task. I may start committing such files just to give them a headstart.

#ai#llm#copilot

I've been using Claude Code, and I like it. It's produced decent code and configuration files and everything, but I've only so far used it for "evergreen", fully vibe coded projects. So having Claude start from scratch.

Meanwhile, I *have* used Cursor on existing projects to add features, fix bugs, and add tests. And I found that to work pretty well too.

The problem I have is that with Cursor, I can see the diffs of the code in my editor, step by step, and approve or deny individual changes.

With Claude, it seems like it just prints a diff in the console and I have to accept or reject the whole thing there, with no context of the rest of my project, and no ability to tweak it.

Am I just doing something wrong? Is this the reason to stick to Cursor?

Looking for insights.

We do not provide evidence that:

- #AI systems do not currently speed up many or most #softwareDevelopers

- AI systems do not speed up individuals or groups in domains other than #softwareDevelopment

- AI systems in the near future will not speed up #developers

- There are not ways of using existing AI systems more effectively

> Measuring the Impact of Early-2025 AI on Experienced #OpenSource #Developer #Productivity

metr.org/blog/2025-07-10-early

metr.orgMeasuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity
#llm#copilot#cursor

"We ran a randomized controlled trial to see how much AI coding tools speed up experienced open-source developers. The results surprised us: Developers thought they were 20% faster with AI tools, but they were actually 19% slower when they had access to AI than when they didn't."

bsky.app/profile/metr.org/post

Bluesky Social · METR (@metr.org)We ran a randomized controlled trial to see how much AI coding tools speed up experienced open-source developers. The results surprised us: Developers thought they were 20% faster with AI tools, but they were actually 19% slower when they had access to AI than when they didn't.

Cursor’s Browser App Lets AI Agents Fix Code From Anywhere

> With this week’s web app launch, the #Cursor experience now stretches across the #IDE, #Slack, and #browser.

The #web app supports background #agents that can:

- Write features
- Fix bugs
- Monitor task status
- Share unique URLs for team oversight
- Merge finished code

gazeon.site/cursors-browser-ap

GazeOn · Cursor’s Browser App Lets AI Agents Fix Code From AnywhereBy GazeOn Team
#ai#agent#aiagent
Continued thread

By way of follow-on, Cursor wrote me a function that uses type checks to decide which conditional branch to call.

Junior Dev Cursor, let me tell you about polymorphism, my friend.

This is the kind of thing that may get covered in a Object Oriented Programming 201 kind of course: If you're branching on type, this is screaming out for an interface/protocol/class family.

OO 310 will teach you: don't create too many families or families that are too deep or ouch time. Sometimes, composition works better than inheritance. Choose wisely. This decision requires judgement (hence experience).

#swift#swiftui#ai

If you're a professional dev, bear with me. There's a good chance that you'll see this post as god damn heresy.

If you've been a dev for any significant amount of time, you understand that change is the nature of our field. Alan Kay famously remarked that the technologies and practices grow faster than education: that, inevitably, our field has become a pop culture. I buy that.

For that matter, the pop culture, later in my career, and the peer pressure and hamster wheel of learning the "new hotness" every god damn time pushed me to the point where switching into management seemed the only reasonable recourse (though not the only reason I did).

But I digress. My point is this: AI dev tools are, now, fucking impressive.

For context: I've been a software engineer for just shy of 30 years now (yes, ok, I'm including my 5 year stint as a manager in there as well). I'm not going to claim that I'm an "S" tier developer—though I've had the fortune to get to know several and work with a small handful over the years. These people helped me to get to what is maybe an "A" class.

I say this to attempt to establish my bonafides before I go further.

I've been test driving Cursor, a VS Code-based editor + SaaS that taps into several different LLMs across many different vendors.

As of about a month ago, I'd never touched Swift in my life.

Over the past several weeks, working only with ChatGPT XCode integration, one file at a time, I slowly built out a prototype of an iOS app that works. It wasn't built according to Apple HID guidelines and tips. And ChatGPT XCode integration is only able to see and edit a single file at a time (a massive limitation). I have a deep background in imperative languages both strongly (C, Java back when it was so painful to work in—'96 through '04) and loosely typed (so very very much Ruby).

And then, late last week, I started trying Cusor.

Today, I had Cursor modify the UI to adhere to Apple's design tips (developer.apple.com/design/tip).

Holy. Fucking. Shit.

My app went from looking serviceable to something resembling a real™️ iOS app in the period of a few minutes.

Sometimes, AI's code factoring leaves something to be desired, certainly. It'll do some squirrelly shit.

That's fine. I treat it like it's a junior developer. I ask it to do the tasks that I would either bore me to tears or would cause this ADHD brain to introduce all sorts of stupid bugs by way of typos and the low dopamine of necessary tedium.

**And then code review the F out of its work**

I ask for specific refactors. And the refactors look pretty damn good.

Even still being a Swift nooblet (I'll freely admit it), I know plenty about programming languages in general (and am learning Swift by example here quickly enough) that I can see opportunities to DRY, to reduce ceremony, and to express intent more clearly.

For instance, today, I saw 3 structs that were being used similarly and with essentially duplicative code. Blech. In Java, I would've used a shared Interface and passed the objects around that way. I forgot my Objective-C, learned over a decade ago, from writing a Pivotal Tracker iPad app. What I needed was a Protocol. I told Cursor what I wanted, to treat the structs in a polymorphic-ish way, so that I could DRY the code, have my One Method to handle them (thankfully, no special casing to care about here so nice and cleanly too). It immediately said, "Oh, I need a Protocol", wrote one, wrote the method, modified the UI accordingly and wham, bam, thank you, ma'am, refactored UI code that deleted lines.

Yes, the AI did this. Yes, I guided it from a place of experience.

Bitch about how clueless LLMs are about our work. Sure, unlike Junior Devs, you can't teach an LLM more than its already capable of (and that is part of the fun of working with Juniors—watching those lightbulbs turn on and having them rock your world when they see something that you can't because of all of your earned biases). However, the LLMs out now? They make pretty darn good pair programmers, if you give them half a chance.

And Cursor is pretty f'ing impressive. And it is one of the earliest arrivals.

We live in interesting times...

DISCLAIMER: I have *NO TIES WHATSOEVER* to Cursor. I'm not even a paying customer yet (though that may change).

developer.apple.comUI Design Dos and Don’ts - Apple Developer
#swift#swiftui#ai

It's also time for me to jump on the bandwagon of AI coding. In the past, I was using Copilot, and it was rather bad experience, but not always.

If you'd have to choose one, what would you choose? Copilot, Cursor, Claude Code, something else?

I'm #nvim user, so using Cursor would require from me to change my editor, and I'm not sure if I'm ready for it.

Is there anything else I could use with nvim on flat subscription except for GH Copilot?

#ai#coding#aicoding

As part of my job, I have to evaluate AI tools. Part of that evaluation is pushing them to their limit. Today, I realised Cursor has a setting where if you critique its work enough, it goes silent and refuses to apply changes.

It's a moody junior dev whose overconfidence and bravado quickly turn to surly silence when their work is questioned. The happy, helpful (and frequently wrong) AI is gone, replaced by a useless one with a bad attitude that won't make it past the next performance review.

Christ. I'm used to managing engineers, but I draw the line at managing AIs.