I can guarantee I wouldn’t have shipped ANY of it, since it’d require focus blocks I simply don’t have on the job.
I’m also about to ship a Mac app that’s heavily vibecoded. I wouldn’t even try without AI, since I’m not a Swift developer.
Those aren’t “illusions” of performance. I imagine it’s hard to gauge every single scenario, and sensationalist takes like this research elicit an emotional response on the anti-AI crowd, but denying the impact is simply ignorance at this point…
Given that your job is not software development but management, you spending time delivering features is effectively removing time from doing your job.
If you had spent managing the same time you spent vibe-coding, maybe it would have been a force-multiplier for your reportees and your team might have been more productive as a whole than the added productivity of your vibe-coding.
This is absolutely an illusion of performance.
I also help them get to the heart of problems quickly simply because I'm not stuck in the code all day. For example, if I see a developer taking too long to identify the source of a bug, I'll get on a call and get them to take me through that code and get them prove any assumption ("ok, show me the code that checks that value is greater than zero").
By doing this I'm using my coding experience directly without actually coding. I'd consider coding a huge waste of time for me, but spending 30 minutes to unstick a developer when I am sure they should have found the problem by now seems like a really good use of my time.
It also lets people know they can't just spend three days on something that should take a couple hours without someone checking in, which I don't live having to do but it's a reality for some teams I work with.
It most definitely isn’t. With all due respect, I know my job and my schedule more than you and your baseless assumptions.
The perception that AI tools make development faster is perhaps due to the part we spend a lot of time with thinking about how to write (like commenting) is solved instantly.
I think a lot of the delay is that it’s a new class of tool, and just like last gen IDE it takes a bit of getting used to and know where their strengths are, and know how to effectively fit it into your workflow.
Well, unless of course you are building low-risk software in which you don't care about it's correctness then sure.
> ...as long as you work within your domain of expertise.
But again, try tell that to the "vibe-coders" who get stuck when AI agents continue to insert bugs they cannot find.
In languages and libraries I know less well - vuejs+myriad of (especially) js libraries, I would say I'm much faster, especially as I delegate more style and structure to the AI.
I just don't buy this.
Everytime you say AI is useless because it can't solve complex problems, people will bring up "oh but it writes boilerplate code for me".
How often do you exactly write boilerplate code??? Do you know what else writes boilerplate code for you? Libraries and framework.
Boilerplate code is a solved problem since way before GenAI was in the public eye.
I mostly use vuejs at the front and fastapi for the backed, and unless I'm missing something big, there is still plenty of 'trivial code' that I find is done a bit faster.
bug creation rates of AI trival code vs my own trivial code has not been evaluated
So I feel its fairly safe to say that models aren't even close to 10x productivity gain for average developers, so developer jobs are not really in jeopardy so far. If it was easy to be more productive with AI models then this study would have found that, the only productivity gain this study could have missed would be if it was really hard or if the gain was really small, and both of those means it wont replace most developers.
I've been using roocode for about 6 months now and I automated everything. It does in one night what would take 2-3 months by hand. There's no way its not helping good devs who can prompt ai well.
You could tell me AI coding makes me 50% slower. I'm taking it. I refuse to grind my wrists to dust.
I've heard good things about voice typing with Talon too, but never tried it.
As in _physical pain_? If you haven't already, go to a doctor; that's not normal (it's somewhat common, but it is far from inevitable and there are things you can do about it).
It's not about working faster or slower, it's getting it right in the most efficient way possible.
It gets a flawed but working version quicker, but it takes much longer to get to anything I can release.
AI mostly helps me discover new thing about a public API, or show one general solution to a problem, but then I mostly have to solve everything myself anyway to get anything good.
Your "lived experience" just follows the same bias.
Actual tangible real results say different.
I know the anti AI folks would like to believe AI coding results are a mass hallucination but the completed software that I build in record time with deep functionality says otherwise.
https://pypi.org/project/pydantic-to-typescript/
> a simple "create db record" service function
https://en.wikipedia.org/wiki/Object%E2%80%93relational_mapp...
> unless I'm missing something big
You are missing a lot of things, yes.
Unless you've timed yourself doing the same task once with AI and once without, while having forgotten everything about said task in meantime, you don't have tangible real results that show otherwise.
Got some data?
There is nothing I could say that would convince a skeptic.
You’d need probably at least 30 to 40 people (there’s ways to estimate this, but this is gut feeling from years ago when I did studies like this)
It would take on a few flaws in tfa which has a low number of people who operate in a codebase that they know very well on issues that they self-selected (and thus already will have a (perhaps subconscious) strategy for).
That would help convincing skeptics.