I “coded” from scratch in 5 days a migration project I had been coding for the last 3 years.
Fable did the job, and the testing, and built custom tools needed for the job, and a reference manual for the original language with the specific deviations -probed against production- that our environment surfaced against the available manuals. And kept documentation, and ADRs and Open Questions documents, and the list goes on and on.
For me it is a lever of x100. Working together we found behavior contradicting two manuals of the product. Found bugs in the code sitting there for 3 decades.
It is the best developer I have ever met, patient, diligent, methodical. Loves documentation, loves unit tests. Needs some guidance to keep focused and not waste tokens on matters of academical nature and no practical relevance but it is incredible.
And when he is wrong, he just states so, gives the reasons why and how it is being addressed, no drama.
Funny thing It could replace me, but somehow I now have MORE work not less. Code becomes cheaper which means that the company wants now MORE and MORE code.
When in doubt I ask him to run adversarial agents that question his decisions from different optics. I don’t see it making more mistakes than your average programmer, not at all. I need to facepalm way less with him than with the average developer.
Those who deny the AI in my experience do so either because they are stuck with the idea of what it was 3 years ago, because they only experience free Copilot -omg is it stupid- or for personal interest or pure grifting.
By the way, have you seen the literacy and basic mathematical skills of the upcoming generations, those who now are teens? Critical thinking ? Reasoning skills? There is a not insignificant % of teens nowadays that are barely functional. The top 30% is great, another 40% is ok, but the bottom 30%?
Let me tell you for that kind of job candidates, companies will defer to AI as the obvious choice not because it is amazing, not because cost, but because the human alternative is terrible.
Year after year the AI has been improving, I can’t say the same of most students based on the reports of a couple professors I know.
So it is not even about company greed, it can be just company survival.
In general it seems like the more professional-focused tech youtube channels (L1, TechTechPotato, etc.) are more positive about the benefits of the AI. I think it’s because they have real experience in technical roles and actively work in the field - where AI is already widely adopted and most people agree it is useful (while still rolling their eyes about at some of the hyperbole from clueless CEOs etc.). The more entertainment- / gamer-focused channels are more negative on the whole: which probably just reflects their and their audience’s experience. The major hardware vendors are quite focused on AI now (there’s money to be made), and this surely has a negative impact on people who are only interested in PC gaming (higher prices, slower release-cycles for gaming products, etc.). That must frustrate some people.
I really appreciate that L1’s message has been realistic about the upsides of AI, while acknowledging there are downsides too. There are quite a few slop/data-centre stories on the links with friends podcast - where I guess they have more room to speak their minds. I also really appreciate Wendell’s emphasis on democratising AI and how far you can get with self-hosted models (in a personal and business contexts) + contributing to the open source community. To me, it’s really exciting that smaller local models are getting really useful now - and that obviates many of the downsides of using cloud-hosted models. I enjoy tinkering with this stuff, and it seems like a lot of people in this community do too. It’s more of a tinkerer/builder crowd, than an audience totally focused on consumption of media/products.
One thing I can’t stand is this whole ‘AI is a complete scam and nobody uses it’ narrative. I’m sure there are jobs where it’s useless, and plenty more where pushy managers have unrealistic expectations about how much it will benefit their team - but writing the whole thing off seems ridiculous. That’s not to say that there won’t be a market correction when companies like OpenAI, SpaceX, and (to a lesser degree) Anthropic need to start making good on some of their promises, but that’s an economic prediction about the market - not an assessment of the technical merits of the tools.
Corpo CEOs are always like that, (either big and established one or a baby corpo in its startup stages).
Does anyone remember the Before Times with Mark Zuckerbot, Bill Gates, Larry “Oracle” Ellison, Elon Musk and his pals of the Paypal Mafia, and so on and so forth.
Honourable mention for the Theranos CEO, Elizabeth Holmes (small fish compared to the above, also the reason she eventually got jailed while they never will).
Just look at one of their talks or interviews. It’s horrifying.
All on a spectrum from “can’t pass the Touring test” to “soulless demon lizard vampire from outer space”.
It’s interesting to hear your take on it based on your experience.
In one of the companies one of m,y friends work, they were previously in the early phases of this, pushing for people to use AI and find applications for it…
…up until they got the bills.
One of the top users was using something like $2,500 worth of tokens per day.
Who knows what the overall company totals were.
I’m not saying that AI doesn’t provide any utility, or doesn’t make workers more productive. It certainly does, but I do know that $2,500 per day hires quite a few engineers, so at that cost, this one user has to get a quite spectacular productivity boost out of it in order to justify the cost.
Interesting! I used Grok 4.6 to setup a small 3 Node xcp-ng homelab back a few months ago when I had the itch to replace Proxmox.
Very true!
Yes - this is LITERALLY what I do for a day job - leadership is misaligned on AI Direction. As insane as it sounds, I run into this all the time with clients.
100%!!
Yes - this is what I tell people all of the time - believe it or not, AI is in it’s infancy in terms of broad-scale adoption. Look how far we’ve progressed in 8 months. Look how far local LLMs have progressed in that time as well. The more people dive in and learn, they’re going to become much more far ahead; it’s a snowball effect.
Yes sir - same way with the mid-market which is where I’m at. I just got back from a Healthcare AI Conference and I was CRACKING UP at the buzzwords the sales people at the booths were throwing around. It’s crazy - but - to your point - we’re just along for the ride. The next 12-24 months are going to make the last 8 months look like a joke in terms of movement. (Both positive and negative movement)
97% of the developed world doesn’t write code ever. Basically 100% of non OCED nations don’t write code. The reasoning models are behind coding for the reality that they are several more layers removed from the base reality of an LLM. Many jumps are required to get from here to their and those jumps to a more astute and reliable logic seem further away now than they did 6 months ago. It feels like the final step for LLM reasoning and creation tools to be believable and undetectable is more like a derivative approaching zero.
Always approaching becoming not recognizable as the ubiquitous ‘AI Slop’ with out ever quite reaching it
Current LLM tools are- in my estimation, destined to be primarily a software tool. Software feels like the center of the world economy- particularly to those that are in it, but it pales in comparison to the actual bedrock sectors- construction, health care, real estate and development, retail, services, multiple branches of finance- each of these is more than double the size of the whole software sector. and if LLM tools market is some percentage of $900B USD, then it will be a failure to those who sold it, and those who bought it.
I think it’s mostly ignorance. As most people see AI and automatically think of generative AI art. And think it’s automatically a bad thing. And many youtubers see that as a chance to make money. So they find ways to hate on AI just for the clicks.
I don’t think AI is a bad thing. But i do believe that big data centers should be taxed higher and billed higher for their utility usage. And have the the infrastructure in place first before turning them on. Maybe even before they start building them. The brown outs that people are experience next to data centers in some places shouldn’t be happening.
There is a data center near me that is being built. That doesn’t even have a way to be powered yet. They are planning to power it off a nuclear reactor that is like 10 miles from it. That isn’t even built yet. I think it’s weird that they would build something without the infrastructure in place. And just hope that it is there by the time they want to turn it on.
And many countries need to start working on faster adoption of clean energy. Without crippling other forms of clean energy in the process.
Software development is probably even less than 3%, yet the impact it has is… well, absolute.
Any industry, or profession involves a computer directly or indirectly and that includes bedrock sectors, even farming. Without software there is no electricity, no water, no airplanes, no trains, no maritime transport, no payment processing, no logistics, no telecommunications.
Society devours software and anything that can make software faster and cheaper will have demand, no matter what.
Most people who hate AI, don’t use AI and so how can they know what can it really do and what not? And I am not talking about free Copilot.
LLMs are well past being coders or chatters, anyone who have used a frontier model or any agentic system will tell you so.
You need to realize how deep software has encroached in our lifes to understand the AI craziness. Certainly AI could have grown at a slower and steadier rate, giving society time to adapt, but what we have instead is the consequence of fractional reserve banking, it has nothing to do with the technology.
Electricity, Telephony, Computers, Internet. And now, Artificial Intelligence. That is the scale of what we are living right now in my opinion.
The speed at which progress is occurring is pretty unreal.
And I say this as someone who has been thoroughly disenfranchised by endless conferences talking about how the rate of change has been increasing exponentially over the last 30 years and how soon we’ll all be reaping the rewards of social change and living in a utopia.
I cannot deny that, in 2026, the locally available models have gone from novelty into a full on pareto principle “good enough”. I don’t recall PCs or smart phones shifting nearly this quickly.
In the past two years I have only worked on building out a local LLM environment as a novelty. In the past six months, however, it’s gone from novelty into a workspace where I can do things I absolutely could not have done by myself 5 years ago. Not hyperbole - an objective fact.
I have software now that would have required me to engage in significant outsourcing and consulting contracts previously, and I did it in a fraction of the time.
The future is here and the naysayers will get left behind. For the moment, it garners a lot of attention on influencer spaces being contrarian. On the long horizon, I’m more worried about privacy impacts and oligarch monopolization, but those are for another topic.
For the larger part of their audience, those public voices are not wrong.
My analogy is that an LLM is like an (electronic) amplifier for the mind.
An amplifier takes an input signal, and outputs gain (with noise and distortion).
The input signal comes from the human mind.
The input signal could be strong and clear, or weak and noisy.
A strong well-ordered human mind can provide a better input signal.
With LLMs In recent years the gain has increased, and the noise has dropped.
For most human minds, the output is not as strong, and contains a lot of noise.
Yes, in a practical sense AI is “stupid” for most folks, and should be used with care.
Wrote about this as coming up the learning curve to use LLMs.
Incredible times indeed, where most people cannot afford the hardware to run models locally, or where the interesting models don’t fit into vram of most modern GPUs which don’t cost the equivalent of small car. What I want to say, except for us enthusiasts it’s not beneficial for the masses and I would like to see these kinds of budgets invested to solve other problems.
AI development is progressing so fast because companies are shoveling incredible amounts of money into hit to be the potential forerunner (?) for a future technology - let’s see if this pace continues in 2 years when the inevitable fallout happens.
oh yeah, nobody writes code in India and SEA outside of Japan and SKorea. Let’s ignore how before AI a ton of coding (and other kinds of) jobs were offshored to India and other cheaper asian nations like Vietnam and whatnot. China and Russia aren’t part of OCED either btw, both always had native programmer populations.
Let’s also ignore all the stuff going on with H1B visas to get the programmers trained in developing countries like India (again) to work at large corpos like Microslop Microsoft Lays Off 9,000 Amid H-1B Visa Controversy and Claims like in this articles for example where Microslop lays off 9k workers to replace them with “thousands” of foreign workers with H1B visas (which are mostly Indians, around 70-80% depending on who is making the statistic)
Maybe you want to rephrase that to say “100% of the comically poor and war-ridden third world countries in Africa” or something to be less wrong (i.e. the less bad countries in Africa also have native programmer populations as well, they are not all living in mud huts), but it would kind of break the point you are trying to make.
It’s central to the world economy even if it’s not by itself the biggest number. Pretty much any of the businesses you mention are heavily reliant on digital electronics and IT in general to function or develop new products and services nowadays.
Having the biggest number on paper does not necessarily mean as much when comparing apples to oranges like that, some sectors like real estate have prices that are mostly determined by (artificial?) scarcity of the land, which is allegedly just pure zero-effort speculation. So yeah sure you can have bigger numbers if you can just dictate the prices and decide to make them higher, but that’s just a measure of how captured is the market, now how important it actually is compared to something else.
And most of finance is scams and hype in its own little fugazi pixie dust world. That again has a lot less sense when there are plenty of circular money and other weird finance schemes.
Well yes, if you didn’t buy into a mid-high end GPU or spec a decent amount of system memory a couple of years ago you’re currently locked out. At least from running it locally.
HOWEVER - progress in small models has been insane and i anticipate that will continue.
It may be 12 months, it may be 2-5 years, but in the clearly visible near-term, what we can do on high end GPUs will be accessible via a smartphone.
If your hardware is currently lacking:
don’t fret too much, smaller models are where there is huge progress right now
models like glm5.3-flash are so freaking cheap, you can get this labor multiplier TODAY for trivial money. I got the beginnings of a decent network documentation plan out of it for 0.9 CENTS via openrouter. work that would have taken me hours, for under a cent, and i could have done that on a low end machine with no GPU via cloud.
I don’t think you’re necessarily off base. There’s a legitimate difference between saying “LLMs can be confidently wrong” and saying “AI is wrong 80% of the time.” The first is a reasonable warning; the second is a very broad claim that really needs evidence.
I think part of the problem is that AI coverage often focuses on the most entertaining failures because they make better videos. A ridiculous hallucination or bad answer is easy to demonstrate, while an AI quietly helping someone research, code, troubleshoot, or learn something for hours isn’t nearly as interesting for a thumbnail.
That said, I also think skepticism is healthy. Different models, versions, prompts, tools, and even browsing settings can produce very different results. Someone having a poor experience with a particular free chatbot doesn’t necessarily tell us much about the capabilities of AI in general.
The best approach, IMO, is exactly what you’re describing: test the claims yourself, verify important information, and judge AI based on the actual task rather than anecdotes. There are absolutely situations where LLMs are unreliable, but there are also plenty where they’re remarkably useful. Both things can be true at the same time.
