The substack popup to join and follow the author made me giggle: Why would I join that substack to read that "author" when I already have multiple free AIs that can give me the same take, but without all the waffle?
In the past, I might have signed up to read the author (I have signed up to a few authors before), but if all the author has to offer is basically already available to me for free, why would I join up?
"It built a machine that turns valuable things into free things, and then pointed it at the businesses it needed to stay alive. That isn’t a revolution."
It is a revolution. Free models are out there. The "toll booths" (software companies) might not survive, but if that's the case, that is the revolution: Because something that you had to pay a massive "toll" for is now essentially free. That might not help the AI companies but it will be a revolution for everyone else that had to pay the tolls, likely making everything cheaper and everyone better off.
(Yes, everyone. Even a very poor person in a highly developed European country lives better than a mildly rich person before the industrial revolution, and I'd say the average person lives at the levels even royalty could barely dream of.)
Even if this bites the AI labs and they don't survive, building software is not going to become significantly more expensive than it currently is with free models that are being sold at cost (barring something major like a global nuclear war), mainly because you can always continue to run the existing models on existing hardware even if progress were to magically stop.
Example? "Massive" is relative, and it's almost never actually massive to the person paying. Software is incredibly cheap.
Adobe is primarily a professional program and even at higher fees it's a miniscule amount compared to the rate of the professional using it. Adobe isn't scooping up 20% of their earnings, it's more on the order of 0.07%.
The entities seemingly paying "massive tolls" are large corporations, who pay the big bucks because they're paying per seat, employing tens or hundreds of thousands of people. The toll per-person is, again, absolutely tiny and nothing compared to the salaries paid.
So in exchange for devaluing the labor of anyone in the firing lines of AI, removing a path to building a business with time, smarts and energy instead of capital, not actually fixing the lower and middle classes main categories of expenses (food,housing,healthcare) where is it going to make everyone better off?
I feel like anyone optimistic about these changes hasn't actually thought it through..
I feel that that is an optimistic take - for now the professional using it can command high salaries. It will not remain that way. The endgame for AI in an economy is to drive the value of labour and knowledge to zero.
As existing models become out of date, somebody has to pay the massive cost for training updated models to replace them, including, of course, the free models.
I need to fill in timesheets, which uses a proprietary API. Claude guessed and iterated until it found out how to use that API, and filling in those sheets is now much less bothersome. I only need to enter a password every three days instead of four times per day.
You don't rely on the model knowing about such things up front. You feed it the latest API documentation and then work off that.
You could simply try it. Find an API that is not publicly documented and thus not part of any training set. Take any AI model, give it the documentation and tell it to create some code that uses that API.
They linked some internal llm with all the confluence documentation sets
My product owner tried to find information about our product, as a regular user would, so feel the experience from a user's perspective
It was a failure: only random, generic answers were received
Then, he retried, specifically citing our main confluence pages: it was a success
Conclusion: for the users that already know were is the documentation, the tool is helpful. Reading the page would be also helpful.
Conclusion (reworded): if you know where is the treasure, then the magic gold-detector will work;
Top models can search the web and read the API docs.
Opus 5.5, for example, can write zig code very well, even though there is very little zig code repos out there.
Training models (of same size) gets more efficient over time from combination of software, hardware, and infrastructure improvements, so retraining that open model to update it, will be cheaper and cheaper.
Related, for the CHM's IBM 1620 project, we received 4 entire cabinets filled with punched cards from Indiana University -- a prof there took every useful IBM 1620 program, and immediately made a copy and put those cards into what ended up being 4 card cabinets -- those deep things, with decks on both sides of each drawer. About 10' x 5' tall x 6' deep. As part of the project, we had a card reader that was capable of reading 'column binary' -- that is, it could read the bits on the cards, not just the encodings. I wrote a program to read ALL of those card decks in column binary and then took those around 200 files and put them into a single .ZIP file.
That .ZIP file holding ALL of the known commercial / scientific computing SW for the IBM 1620 fit onto exactly .... ONE standard 1.44 MB Floppy Diskette.
Yesterday’s models produced okay small results, but unmaintainable large results. People bet that models would continue to improve, so that tomorrow’s model can fix today’s slop. This is a gamble, and it’s not clear yet whether it will pay off. If that progress stops for any reason, you end up stuck with a model that’s good at starting things, but not good at maintaining them, and a bunch of code that gets more and more expensive to maintain. https://maevyn.codes/decoherence is worth a read.
And that’s quite apart from the inherent decay due to information/reality drift.
There are model that some people are paying for and choose to let you use gratis because that brings them value at the moment (eg: it hurts the competition);
Once/when the war is over, nobody will keep paying billions per year pro bono;
I'm skeptical. It's not easy to capture all of the behavior of piece of professional software that has evolved for decades. There are already dozens of cheaper (and free) less-capable photoshop clones that haven't taken over the market.
I suspect the bigger issue with Adobe's business is that generative AI is quickly replacing the graphic artist. The replacement for Photoshop isn't PhotoCraft, it's a prompt.
The artcraft versions are rough, but good enough for what I do. Mostly background removal, blemish removal, exposure and white balance adjustments. A professional’s mileage will almost certainly vary.
I’ve tried Gimp, photomator, a few others. The artcraft versions at least retain the muscle memory of the creative suite apps, whereas the alternatives all feel like I need to start from scratch to relearn their features, limitations, shortcuts, and quirks.
If Adobe had frozen features and offered a paid perpetual license pinned at a certain version I probably would’ve bought it.
I think that’s one of the things that’s jumped out at me in this latest slop wave from the last few weeks. Part of releasing meaningful software is the time investment by the creator is at least the barest of assurances that it’ll get some kind of support going forwards. Obviously not always, but the bigger the scope the more likely that is I think. If somebody spends 6 months on a project then it means it was worth at least 6 months to them and I believe there’s a higher chance they will continue to support it.
With stuff like this, it’s unusable and unmaintainable, and it got crapped out so fast that the creator has absolutely no emotional or personal investment in it beyond the token costs. I’m not suggesting every project needs a minimum amount of time spent on it, but it’s certainly a factor in how I judge software now, and choose what I want to interact with.
I have similar feelings about getting AI-generated "communications" from people (email, whatsapp, whatever).
The value of having a human send me a message is that there is also a signal in there - the human at least understands what whey they sent me. They may misunderstand the concept, they may be mostly inarticulate, they may in fact be altogether wrong, but at least I know what they think, and I know when to ask them for clarification.
When a message is sent from an AI, I have no idea if the human doing the copy-and-paste even understands what they are sending me. The signal that they understand the contents of the message, that signal is missing.
It's no different to asking someone a question, and them pasting parts of a manpage as their entire response: I can't tell if they understood what they sent me.
I think this is one of the biggest open questions, not only for cloning software but for "vibe coding" in general. How hard is it actually to turn such things into polished products? Is it even feasiable or are you just drowning in slop at some point?
No, someone still has to pay the subscription. AI Labs don't care who uses the tokens and how. If Adobe goes under someone else is going to take their place. Suprise, surprise: they will also use tokens.
> It’s a very expensive way of shooting yourself in the foot and sending the bill to your investors. <CoreWeave bonds chart>
I'm not sure how CoreWeave bond chart is even remotely relevant here. They are not even an investor.
> They are just mining the asteroid, dumping the silver onto the market for $20, and watching the price of their own product plummet to zero.
The lab's own product is the model. How is someone building a weekend project going to plummet the value of their model? It's the opposite!
Some dev or company might pay for a subscription (or self-host an open weight model) for maintaining that Photoshop clone, but the millions of people that were paying for Photoshop could simply stop paying and use the free/cheap alternative.
No one will be able to pay for tokens if their product is replaced by OSS created by open weight models.
True, with one exception: they do care if people are using the tokens to distill their own models. There is just not that much they can do about it except for clumsy heuristic type protections that will never really work.
Even if AI labs don't really care if they are getting paid by companies or individuals currently, they do have two pretty big problems to worry about.
One is competition from "open" models and just the complete lack of any moat for anyone this many years into things, and the second is that if all the software developers who are writing vibe coded clones of commercial software actually do lose their day jobs because the software companies they worked for go under (and if there are no more jobs for them to go to because this is happening everywhere), they eventually won't be able to afford the big lab subscriptions for personal use, so both pools of customers for the AI labs collapse in a related way.
I do think the linked substack post has some serious deficiencies in its analysis but at the same time it at least starts to sniff out that commoditizing knowledge collapses everything for everyone in ways that will be incredibly destructive to our current economic system.
Someone stealing from the stealing machine, oh that irony. Also hard to tell if they really have such a huge problem with it as that token spend makes their numbers go up significantly.
> One is competition from "open" models and just the complete lack of any moat for anyone this many years into things, and the second is that if all the software developers who are writing vibe coded clones of commercial software actually do lose their day jobs because the software companies they worked for go under (and if there are no more jobs for them to go to because this is happening everywhere), they eventually won't be able to afford the big lab subscriptions for personal use, so both pools of customers for the AI labs collapse in a related way.
If they hike up prices they kill demand, either because they compete with humans or open weight ... and now it's both actually. If they don't hike up prices they can't pay the bills for their absurd capex and they kill demand long term because there will be no one left to pay for it. Pretty fucked up situation I would say.
They are banking that a bigger model can get distilled (internally) into a more capable smaller model as opposed to training that model without having access to the bigger model in the first place. Majority of the value generation seems to be happening at the frontier though, not in smaller models, and I'm not sure that gap will shrink any time soon.
> all engineers lose their jobs
They want to replace engineers with agent swarms. At that point it stops being a tool and they can charge for outcome. They might get there, or we might still need "qualified steerers" in the future. Who knows
What keeps Adobe up at night, for better or for worse, is not that somebody is going to make a product that imitates what Photoshop used to be but that somebody is going to come out with a much better tool for AI driven editing and generated images.
Even pre-AI, I or other team member could have written our own versions of:
- JIRA
- Prometheus
- etc
but we didn't.
Why not? A couple reasons:
- Why write custom software for common problems that lots of other people have? We are already helping build internal custom systems for the business
- We were already supporting internal apps and now we have to support more apps? Apps that have dedicated support teams that are cheaper than us supporting them due to economies of scale/
You might say that AI changes the economics of the above to which I respond:
the marginal cost of a vendor adding another cloud instance of their existing software is, while non-zero, MUCH cheaper for them than for us. They also handle all of the security, patching, stability etc.
Yes, AI changes the absolute prices of this work but it doesn't necessarily change the relative pricing dynamics which are effectively the same.
If we assume that this is indeed true, that would mean AI companies will go bankrupt, leaving a lot of hardware to be rented for cheap. Isn’t it something that would benefit the community?
Or maybe they’ll find the next vertical and just move there now that “software is solved”
> That isn’t a revolution.
What do you think a revolution is? Revolutions are literally complete upheavals of the system. The problem that you're really identifying is that this revolution appears to be unanticipated by the players fomenting it, and it also seems to be unintentionally targeted directly at themselves.
Wherever this goes, it's sure to be interesting.
Ultimately Photoshop as an idea stopped creatively evolving and Adobe created a tenuous pricing scheme that disincentivized competition, locked in users, and made most customers resentful. What this taught many of us is that this is a cautionary tale.
Mind you Adobe could have made CC free, built a healthy ecosystem around it, and created a lot of value in network effects, standards, plugins, etc. But they chose the dark path and they deserve to pay for it.
For AI labs, it's not bad at all. They make their hey while the sun shines. It's worth it. For Investors, they are gamblers after all. They get their fun in betting. Totally worth it.
Yes, building something as complicated and locked in as Microsoft Word required a lot of work, and Microsoft has gained considerable money and power over it, but that doesn't mean that Word is necessarily good.
No different from "McDonald's hamburgers are objectively very good."
For a while I thought it was annoying that Word wanted to save my documents in the cloud but then I realized I could keep my lifestyle diary for my nutritionist in Word and update it on my iPad on the go and it works great.
That being said let's keep in mind that the latest wave of moving-away-from-MS-Office is using OnlyOffice this time and I believe it might work. OnlyOffice actually wants to entice MS Office users. And thus it is more compatible, has a much closer/better UX, and despite being a web app it is somehow more responsive than LibreOffice.
This is a pretty biased take. If there is anything the world is finally learning, it is that being completely dependent (not only for functionality but also availability) on a single vendor operating from a potentially uncooperative government, is not a great position to be in.
Most of the revenue for Adobe is not due to individuals who need a competent standalone tool. It’s for organizations that need all of the usual qualities:
1) continuity with prior work (my old projects open properly, last year’s training is still relevant)
2) being able to hire/train employees on the tool
3) collaborating externally
4) interoperability across work products (eg vector, raster, video, etc)
You are not getting any of that from someone’s vibe coded weekend project.
Ai models are not up to the task of a complete replacement. Humans are still needed and I don't see this changing. The "news worthy" cases of open source replacement are not in fact usable alternatives, missing features or having serious bugs. Anyone who has pushed on the models and vibe coding knows they have their limits around problem complexity - and that it takes an experienced practitioner in the subject Ai is being applied to in order to get good results.
They are not. CEOs are.
That would be fair, but leadership decides who to cut, so... here's your belongings, please be out of the building by noon.
But, you are moving the goal post.
Instead,
> For three years the pitch from the big AI model labs has been beautifully simple. Software engineers are the most expensive employees on Earth.
The fact that we do proves that ICs are equally to blame, because evidently the majority of them love using AI.
Dude was like “write me a diatribe about AI arguing why coding agents don’t make sense”
Why is there not yet fully open models made by/for the FOSS community (as opposed to open weight by entrepreneur-style companies)? Cost of training? But what is this cost really?
- Do not include costs of running model in cloud - Do not include IP costs - Do not include salaries, marketing, etc. (all these are separate expenses, different depending on individual goals)
There would be pre- and post- training "systems" that could be repeated.
Training data/acquisition could be separated out (so IP-liabilities could be taken on by individuals).
The idea would be a system for small groups/individuals to snap together the different aspects of training (data, pre-training, training compute, post-training) and to train their own (and share).
Won't this soon come down to a few thousand dollars for models that run on modern consumer hardware?