Depends on how in-the-loop you want to be. I personally delegate all the writing of the code to agents, but I maintain a clear mental model of the architecture, which I come up with by iterating and prototyping with agents. I can use Luna for all of this, although I switch depending on the task. It's nice to be able to throw a 1,000 word rough spec at Fable and get a personal tool that works perfectly though. I expect that as the models get better, I'll continue to be in the loop, but operate at increasingly higher levels of abstraction.
I disagree with Musk's politics but it does not impact my decision to use Grok. That's because being serious about aligning my capital to my values doesn't leave much in the way of eligible products or services. I consequently decide not to worry about this as a moral axis for my life.
We've seen plenty of examples of apps successfully vibe coded by non technical people, including apps making real revenue.
Your friend could start with telling the LLM that they are a non technical person who wants to make an app and it will explain all the successive steps.
Maybe people don't cite specific apps because they like their jobs, and outing apps as vibe-coded is still seen as negative
I personally know of two completely vibe-coded large apps in my professional environment. One by a non-technical manager, made to solve his needs, then sold to customers. Initial development went along great, but by now velocity has greatly slowed down. Also took a lot of engineering hours (of actual software developers) to get permission management from "chaotic and ineffective" to passable. It's still worse than what you would have gotten by just using a couple sentences of the right technical language at the start. Deployment is also a bit of a nightmare. All in all, anything beyond the first rollout phase was delayed by months. Honestly it should have stayed as a prototype that then gets rebuilt from the ground up. But still, it is a real app, making real revenue
The other example was vibe-coded by a software engineer in his free time. Works pretty well, doesn't have too many bugs. Makes some revenue, but a lot less. Solving manager problems just sells better.
I vibecoded a Postman/Insomnia API tester program, and I use it everyday at work now.
But as another software engineer, I remove myself from that comparison, because the idea is to find out if a non technical person can do the same, that's the definition of vibe coding.
A friend of mine, non-technical, is not making money with his apps. But he's creating a street fighter like game. Just for fun.
So there's that.
He can't exactly release it because he uses a lot of copyrighted stuff. It's also meant only for himself. Though, I've been asking if I can play it, it looks fun.
I think we desperately need to start differentiating between "is creating" and "has created". I have a couple of "am creating" projects too, but their proximity to "have created" is directly proportional to how much effort and expertise _I_ am bringing, not so much related to the AI's contribution.
I run into this quite a bit. We have users generating MANY apps at our small company (30 FTE), entirely with Claude. It's great to see people mess around and tinker. It's NOT great to see someone with a GH repo that has 750+ commits for what would be MAYBE 1 week of a developers time. SO these are non-developers now spending hours and hours working on software that is probably going to get thrown out.
We're in this spot where we don't know when to cut our losses on projects like this. (Is it even viable as production software? Does it currrently do what it's supposed to, or are they adding new features? Is there a return on continued development efforts?)
None of these apps they have built are seeing any major usage, and I don't think a single one is what I would call "done" (There was a gold rush stage at the beginning of 2026 where senior leadership wanted everyone to spend some time messing around with Claude). Unfortunately, they never told anyone when to stop messing around with Claude, so the ROI is ever diminishing.
The problem with lovable, from someone with insider knowledge, is that many of the apps existed even before appearing there and where ported to the platform to ride the hype wave.
AMD’s software experience is riddled with bugs rendering out of the box training with AMD is impossible. We were hopeful that AMD could emerge as a strong competitor to NVIDIA in training workloads, but, as of today, this is unfortunately not the case. The CUDA moat has yet to be crossed by AMD due to AMD’s weaker-than-expected software Quality Assurance (QA) culture and its challenging out of the box experience.
[snip]
> The only reason we have been able to get AMD performance within 75% of H100/H200 performance is because we have been supported by multiple teams at AMD in fixing numerous AMD software bugs. To get AMD to a usable state with somewhat reasonable performance, a giant ~60 command Dockerfile that builds dependencies from source, hand crafted by an AMD principal engineer, was specifically provided for us
[snip]
> AMD hipBLASLt/rocBLAS’s heuristic model picks the wrong algorithm for most shapes out of the box, which is why so much time-consuming tuning is required by the end user.
etc etc. The whole thing is worth reading.
I'm sure it has (and will continue to) improved since then. I hear good things about the Lemonade team (although I think that is mostly inference?)
That’s insane. There should be a big team of people at AMD whose whole job is just to dogfood their stuff for training like this. Speaking of which, Amazon is in the same boat, I’m constantly surprised that Amazon is not treating improving Inferentia/Trainium software as an uber-priority. (I work at Amazon)
> “Are we afraid of our competitors? No, we’re completely unafraid of our competitors,” said Taylor. “For the most part, because—in the case of Nvidia—they don’t appear to care that much about VR. And in the case of the dollars spent on R&D, they seem to be very happy doing stuff in the car industry, and long may that continue—good luck to them.
Where's the scope for an L7 promo in "Fixed a bunch of tiny issues that were making it hard to use Tranium/Inferentia with PyTorch"?
Amazon's compensation strategy, in which you primarily get a raise years in the future for tricking your management chain into promoting you is definitely bearing its rotten fruit.
Anecdotal but over several years with an AMD GPU in my desktop I've tried multiple times to do real AI work and given up every time with the AMD stack.
Im running fine on my AMD 7800xt 16gb... Yes memory is a bit limited, but apart from the i have found that it works great using Vulcan in LM studio for example.
ROCm works great too, the only issue i have had is that my machine froze a couple of times as it used 100% of the graphics and the OS had nothing left. Since moving to vulcan i stopped getting these errors apart from a little UI slowdown when i had 4 models loaded at the same time taking turns.
Im also on a i7 6700 with 32gb DDR4 so im sure that is causing more slowdowns then the graphics card.
Yet another reason to doubt claims that ”software is solved”.
Anthropic did retire an interview take-home assignment involving optimising inference on exotic hardware, because Claude could one shot a solution, but that was clearly a whiteboard hypothetical instead of a real system with warts, issues and nuance.
i'm doing inference on a free mi300x instance from AMD right now. not sure if the software stack is just old or what, but here's what i've observed: stuck on an old version of vllm pre-Transformers 5 support. it lacks MoE support for qwen3 models. oss-120b is faaaar slower than it should be.
int8 quantization seems like it's almost supported, but not quite. speeds drop to a fraction of full precision speed and the server seems like it intermittently hangs. int4 quantization not supported. fp8 quantization not supported.
again, maybe AMD is just being lazy with what they've provided, but it's not a great look.
right now the fastest smart model i can run is full precision qwen3-32b. with 120 parallel requests (short context) i'm getting PP @ 4500 tokens/sec and TG @ 1300 tokens/sec
From the papers I've read and the labs that I have worked in personally, I would say that most scientists developing Deep learning solutions use CUDA for GPU acceleration
Yeah, historically it’s been software that’s limited AMD here. Not surprised to hear that may still be the issue. NVidia’s biggest edge was really CUDA.
I don’t know what’s a chicken and what’s an egg here. But ROCm support is often missing or experimental even in very basic foundational libraries. They need someone else to double down on using their chips and just break the software support out of the limbo.
This is what I've heard on the "street". Building a CUDA-compatible stack for AMD's hardware requires highly-paid SWEs. It's a very niche field, and talent is hard to come by.
But AMD does not want to pay these specialized SWEs the market rate. Their existing SWEs would be up in arms saying, basically, "what are we, chopped liver??", or so the thinking goes.
So AMD is stuck with a shitty software stack which cannot compete with CUDA.
If I were making such decisions, I would just cull the number of existing SWEs down by 50%, and double the pay for remaining ones. And then go out and hire some top talent to build a good software stack.
As someone who is building an LLM-powered product on the side, using AI coding agents to help with development of said LLM-powered product and for my day job, and has a long-tail of miscellaneous uses for AI, I suspect you're right.
I skimmed through it and it's a simple and concise overview of the language suitable for a beginner. It is by no means exhaustive. Also, haven't seen anything in there that makes it so that it couldn't have been written in 2015
The problem with AI-generated content is not necessarily that it's bad, rather, it's not novel information. To learn something, you must not already know it. If it's AI-generated, the AI already knows it.
We might also say the same thing about spelling and grammar checkers. The difference will be in the quality of oversight of the tool. The "AI generated drivel" has minimum oversight.
Example: I have a huge number of perplexity.ai search/research threads, but the ones I share with my colleagues are a product of selection bias. Some of my threads are quite useless, much like a web search that was a dud. Those do not get shared.
Likewise, if I use LLM to draft passages or even act as something like an overgrown thesaurus, I do find I have to make large changes. But some of the material stays intact. Is it AI, or not AI? It's bit of both. Sometimes my editing is heavyhanded, other times, less so, but in all cases, I checked the output.
How does this technically work? Is it just a natural language shortcut for prepending text to your context window, or does it pull information as needed as inferred from the prompt? E.g. the meeting note formatting "memory" gets retrieved when prompting to summarise meeting notes.
reply