Yep. I've got an "ideas" folder in Apple Notes I've been curating for years - brief ideas I've had while falling asleep and need to jot down, ideas that came to me on the bus or the toilet, something I wish existed, etc.

I've been slowly working through that folder and implementing them. It's been amazing.

It wasn't a concerted "weeks" of effort, it was ideation over the period of weeks until I had a coherent and complete architecture. But thanks for your concern.

You realize that architecting software correctly takes time, right? Often much more than weeks, for any significant effort. You can't just use react and node for everything.

Reasoning is working through a system of logical rules and priorities based on first principles.

Machines absolutely can and do reason, now. Far better than most people, and certainly faster.

Reasoning is merely the exercise of logical rules. The agents instantiate those rules for the transformer models and orchestrate - nothing more.

I don't understand your argument. It's irrational.

I'm basing this off my own adoption and experience. I'm generally an extremely cautious late-to-adopt person, which makes this all the more hilarious to me.

I need to only look at my git commit and feature velocity over the past 12 months to know precisely how useful it's been. Proof is in the pudding. I'm not fooling myself by copypasta'ing things. (I very rarely copypasta things. I design things and my prompts take 20+ minutes to complete.)

I'm shipping actual production product at any increasing rate, and the code and architectural quality is better than most of the people I've worked with in my career (because I'm the gatekeeper).

AI isn't getting more expensive, the ability to use it effectively is getting easier (which increases cost). When you've got 10 claude agents running in concert, or opencode dispatching a dozen subagents to complete a job, it's going to burn through credits. But, that's improving, too: z.ai's GLM and Kimi K2.5 are markedly cheaper than Claude and in some things do much better. Open models are going to destroy the cost disincentive in the next year, I think, and we'll see "failure to adapt" companies fall off (like OpenAI). Everything will ultimately just be weights on open models, and perhaps 1-2 frontier models sticking around.

You're not looking hard enough.

I know a college kid who's in the midst of making a fully open source milling machine and their objective is to make it something that can be made without machine tools (eg. no/minimal machined parts). It currently works and can mill aluminum and mild steel. He's 22. I know of several similar stories.

The local "startup incubator" company that's been in business for 8 years that gets grant money is now in shambles and can't find any customers, because people are able to completely bypass that early stage difficulty in starting a business, thanks to AI (and solo operators doing much the same thing as the incubator, as a turnkey service).

It takes time for tech sector improvements to trickle down into the physical world. You forget, AI is only really useful for about 1.5-2 years, and didn't really hit stride until early December. Every model, including open source ones, released since November has significantly eclipsed what was available in the months prior to November. We're apparently due for another shakeup in models RSN, too.

We sometimes forget how fast things move when looking at a snapshot of where we are. Consider now vs. 5 years ago, or 10 years ago, with regard to tech. Watch an old TV show and notice how the prominence of cell phones simply isn't there. AI has already been consequential in gutting the tech industry (both jobs and salary), and will continue to do so.

With regard to "using AI as a better search engine" - that's huge. We basically didn't have usable search for close to a decade, with the quality of search results becoming increasingly laden with ads and significantly less relevant for technical research (geared towards selling things). That advent in and of itself is going to help everyone, not just "those who know how to search properly" (which hasn't been relevant for years, due to search engines largely ignoring your actual search criteria and giving you slop).

"If you're looking for facts and look around the internet yourself, you usually get a pretty good idea of how trustworthy the source is. "

Until you learn that sites like wikipedia have been actively gatekept by powerful moneyed interests, that "trustworthy" media sources like the New York Times has been run by state intelligence aparatai since the 1940s, and that the cultural indoctrination received in public schools has been carefully manufactured for 70 years.

If your first beliefs are wrong, you're not going to find truth. You'll find confirmation.

It's not just "Chinese made" stuff.

Ever go to the mall and see those kiosks manned by Middle Eastern young people, maybe a pretty girl or a guy, selling these things (or a video game console, or similar things) for under $100? Usually it's around the "fast food for 2" range - cheap enough the average person with money doesn't even think about it. They usually operate around towns with military bases.

They're a well known Israeli Mossad operation to get listening monitoring/devices onto military bases.

You could always try to do something meaningful and original with your work yourself.

Your post denies the status quo. That isn't how any of this works at all.

It's a distillation process. AI absolutely can (and is) being trained on AI-generated, AI-augmented, AI-processed, and AI-sintered data.

I'd suggest you familiarize yourself with where things are today (as opposed to 6 months ago, or 2 years ago). If you haven't reevaluated state of the art in the past 2 months with any depth, you're gravely behind.

To say that open source is being killed by vibe coding is just... crazy. It's simply wrong.

There are numerous vibe coding apps now which were written entirely with vibe coding. An entirely new paradigm of development exists today which didn't exist even a year ago - claude code, opencode (omo/slim), codex, cursor, and on and on. Then you've got the agentic stacks, and everything else that's largely open and free. "Come use this vibe coded thing I built this weekend! It's live in production, you can see it working - and here's the git repo!"

Coding skills are no longer a barrier to iteration and improvement. There are so many cool projects out there now being done by people who have an idea and see a business case and want to fill it.

Aside from the projects, I've already taken 2 libraries myself and forked them to change (and improve) functionality for my specific use cases. I'm assuming they don't want my changes, but they can always pull them back if they want. They can see I've got a fork. My willingness to deal with "well they may not accept my changes" + slowing my own velocity is low. The repo is public, the commit comments are better than anything I've personally done in the past.

"AI and vibe coding breaks that process at the early stages because there is no longer the humans looking at new things and taking a chance on something new and unfinished"

Um... have you even tried vibe coding? You can one-shot a project in 20 minutes. I've done it numerous times - an old project I spent weeks writing specifications for, boom, done. I also now have a very useful data indexer which integrates smb shares with MacOS finder. Any sort of idea can now quickly come to fruition in a couple hours with a good set of prompts. Want to make an antiquated database format convertible to a newer platform, and reimplement the frontend? I once had to take a 15-year old physical SCO system running a proprietary database over to a virtual environment, 10 years back. It was a painful process, because SCO and failing hardware. But today? Once I got to that point I could've reimplemented it anew in a couple days, allowing those companies to expand the software capabilities they paid hundreds of thousands for at the time, to something which suited their current business needs (which were a paper and spreadsheet process).

A mildly capable office tech could take an existing git repo of their project tree and maintain it/add features and maintain the product well enough using vibe coding, instead of languishing for a decade, like they had to previously.

You seem to be missing the fact that LLMs have vastly exceeded prior functionality. Today, the frontier models are easily 2x what they were in October. October was easily 2x what they were in May of last year. May of last year? 2x as capable as they were the year prior. We're approaching exponential improvements, and models have been solving previously-unsolved NP hard problems: that's innovation.

If you have an idea, it can be done with vibe coding today if you have the intelligence and creativity to do it. Simple as. If you don't, you can't - and won't.

Astounding. I can't believe you managed to tie that back around to AGW. Bravo for commitment.

(This is the most unhinged rant I've seen since I last looked at reddit.)

Sounds a lot to me like they're looking for governmental subsidy. They see the end of the runway on what is effectively an unprofitable business model (or at least one they, and OpenAI, have not been able to capitalize on in a meaningful way), and they're looking for a way to "democratize" their costs.

Yep. The default activity it should perform when there's no control is to turn emergency lights on and pull over/park on the shoulder - at a minimum.