Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
No thanks, I'll keep my AI stuff in a web tab and have no appetite to install it on my local machine.
I mean, depends on your work. You don't sound like a developer. As a developer, I've sandboxed mine to specific development environments where I work locally. I snapshot with git as I work and then push once I've reviewed everything, and from there I can deploy to test environments or test devices depending on if I'm doing web development or app development. It's pretty seamless and nothing scary is going on.

--

Ok, I NEED TO RANT.

Before people come at me, I've been forced to do this by the industry. Your proposal won't win if you aren't working this way because you're too expensive otherwise. Ask me how I know. I was more than happy to work the old way before we had to start draining water and polluting everything with giant data centers. This new way is definitely way more efficient, but the problem I'm seeing right now is when everyone is way more efficient, there is less work to go around. What I'm unsure about is whether bigger, more advanced projects will begin filling the gaps more as companies realize their money can go a lot further and therefore they invest in more advanced long-term development work, or if they will be satisfied to get roughly the same amount of work back at a much lower price.

I think we may shift towards more advanced work eventually, but it's kind of a painful time right now. I also don't know how long we have until it all pops, because the growth isn't sustainable, and these companies can't keep subsidizing the price of AI forever. I have 5.6 Sol Ultra and Fable 5 Max and honestly they are good enough for 99% of dev work at this point. So I wouldn't be sad if things just kinda round off on the growth curve and everyone chills out for a while. These are still incredible tools that we could refine to make more efficient that would last us for decades. At current growth rates for AI energy use, the oceans will nearly boil in a few decades. I'm not even talking about classic climate change, I'm talking about simple thermodynamics, you can't just keep adding heat to the system forever and expect it to all dissipate through radiation into space. And don't even get me started on data centers in space, that's the dumbest idea I've ever heard. People think space is cold, and it is, but there are so few atoms in space per cubic meter (we're talking tens) that there is nothing to even dissipate it into. I mean hell, the James Webb Space Telescope out at Lagrange Point 2 operates behind a multi-layered (I think 4?) shade system because the sun makes it overheat otherwise.

There is also a lot of crap code going out right now by amateurs or shady devs trying to make a quick buck. Those projects are going to blow up. Those are the vibe coders. As a senior dev, project manager, and UX/UI designer for the past 15 years, what I'm doing is agentic engineering. There is a big difference. Software slowly begins rotting from the moment you create it. If you don't maintain it, it's all going to fall apart. I'm hoping for a huge tidal wave of work in the next year or two as these subpar dev projects begin blowing up. Many of them already are internally at companies, from what I've read. These companies think you can just build it and be done with it, and they are building at insane levels with less experienced employees than I've ever seen in the history of this industry. Maybe they are hoping by then that they can just use AI to maintain it, but I think for some time, nothing will beat human oversight of these technologies.

I mean, we literally lost a proposal to a business owner's nephew, an actual child, because they said he could just do it in Claude in a few hours. But does that child understand UX? The buyer's journey, how you guide them to buying your product or service? Do they think through every stage of the SEO? What about effective marketing copy targeted at your specific demos? Do they understand the intricacies of balanced typography and establishing hierarchy? Do they even know what blue teaming is, much less the basic best practices to secure your site? Do they know how to setup and conduct unit tests, or what those even are? Do they know the proper database system to use for your stack? Did they even have a meeting with you to talk about your future growth needs, to ensure you're maximizing your investment and won't have to spend a ton for a real dev to retrofit your janky system in two years? How will the site be maintained going forward, when they're away at boy scout camp and your site explodes? Because I can tell you, the AI doesn't just automatically do these things. You have to guide it through the full stack from top to bottom. That's where an experienced developer/designer shines. And we're just an email or text away when your site inevitably explodes. As a matter of fact, we would've setup monitoring and fixed it before you or most of your customers realized it was broken. Did your nephew do that? No? Oh, that sucks. That's really unfortunate for you.

/RANT. Thank you for coming to my Ted X Talk.
 
“Close to magic”? Are you 8 years old? Or 80? Or an AI bot paid to write this blind praise? No wonder these ludicrously overvalued companies can get away this, because they convince old, out of touch decision-makers that their technology is “close to magic”.
The first part of his post is valuable. Not everyone is a tech producer or engineer. If a consumer reacts that way after using a product that means it’s a really, really, good product.

I reacted the same way the first time I experienced active noise canceling earbuds from Bose or the Samsung Foldable display smartphone.
 
  • Like
Reactions: jjrtiger
The first part of his post is valuable. Not everyone is a tech producer or engineer. If a consumer reacts that way after using a product that means it’s a really, really, good product.

I reacted the same way the first time I experienced active noise canceling earbuds from Bose or the Samsung Foldable display smartphone.
Not completely. Today's ML-based algorithms are really good at fooling the user into thinking it is intelligent. What I find, as a developer, is that it can do some things fairly well and at other things it is a complete disaster. It will confidently tell you 2 == 3 even when presented with proof otherwise. I had a recent experience where it asserted X and I told it X was not possible because of Y. It refuted Y was a thing and re-asserted X. I gave it input Z which supported my assertion and it again dismissed it and said X is true. I finally, very forcefully, gave it direct text from documentation and it finally wrote a test, ran it, and said that it had made a mistake. This is not a one-off occurrence either. The best way to describe ML-based agents is like babysitting an L1 who has access to everything but is absolutely incapable of putting it together consistently.

I have more or less been forced to integrate ML into my daily work (we need to justify this expense somehow) and it is so far slowing me down more than helping.
 
  • Love
Reactions: Zwhaler
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.