"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..
> 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.
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.
Models only have information up to the date when they were generated. APIs get updated and expanded, new frameworks and libraries get introduced, bugs get fixed. How long is an existing model going to still be useful? A year or two at most?
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'm using Claude on APIs on which it was not trained, and it's doing well. The techniques it uses to debug and find solutions aren't that different from older APIs. "Let me dump all of this response and guess which field might be what" sometimes, "let me get a stack trace just before it goes wrong", many familiar development techniques.
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.
Yes there are reasons to create better models. Baking in latest revision of assorted API documentation is not it.
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.
Not even human's can keep track with all the latest APIs.
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.
Opus 5.5 can write IBM 1401 Autocoder very well (used on the Computer History Museum's pair of working 1401s). THAT is astounding to me.
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.
Models can ingest up to date documentation, source code and search the web, context windows are big enough these days. Sure at some point retraining will be needed but not every few month and you likely don't have to start at zero every time.
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;
The irony here is that the article, written solely by Claude, is also a volley at authors.
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?
Does anyone actually use PhotoCraft? Is it really a complete replacement for Photoshop or is it basically a proof of concept?
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.
I tried the Lightroom alternative, it was garbage, not even close to a replacement. They have spent decades fine tuning the smallest of features, you’re not going to replace that overnight.
I’m not a hardcore Lightroom/Photoshop user, but I have hundreds of RAW product photos that I need to cycle through a few times a year, and a few new ones to add whenever I make a new product. I long-since canceled my CS subscription because it wasn’t worth hundreds of dollars a year to me.
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 use it everyday- I don't need Photoshop anymore- its faster and does everything I need it to do- the only other "creative" artcraft app I've really tried to use is Filmcraft, the Premiere replacement but I'm still using Davinci Resolve Studio as Filmcraft isn't as far along as Photocraft but Davinci Resolve Studio is slow, bloated, can't customize the UI and barely has an API so I can't wait to dump it
It's not an ideal replacement as of yet, but the writing is in the wall and I bet in a few year time it (or some other clone) will be quite mature and stable.
2) Competitive against Photoshop among professionals.
Until it hits the first point, which it hasn't yet, it's just slop. AI enjoyers don't care how broken it is and will use it anyway just because it's AI generated but everyone else will stay away, especially most pros.
If it can't hit the second point then at best it's just another image editing app, and those already exist. Adobe isn't afraid of Krita or Photopea.
I'm using it daily. It's not as good as Photoshop with brushes, the quick selection and masking features and the type engine is not as sophisticated. But for the simple edits I need to do, it is more than adequate.
Earlier this year we had prospective customers do comparisons against our competitors but also against prompting their own. All of them, without exception, found that to be a bad idea.
If it were just about the ability to edit images, Gimp would have obsoleted Photoshop decades back. Photoshop excels because of its capabilities but even more because of its ease of use and the stickiness of the skillset its users have spent developing.
Nobody uses it. It’s completely broken and was dead on arrival. It can’t even reliably save files. Maybe eventually it’ll get to a sort of working state, but I doubt 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.
On the other hand, we are barely a month in, that thing is still ongoing so it's hard to judge now.
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?
> that money doesn’t migrate to the AI lab. It evaporates.
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!
> No, someone still has to pay the subscription. AI Labs don't care who uses the tokens and how.
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.
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.
> 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.
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
just an aside on the Claude prompter responsible for this article, he labels himself as writing about how "feminism ruins tech", and he has written some pretty heinous things about women. Given that he approaches things with such a (male) software engineer exceptionalist point of view, I would take his writings with a grain of salt.
There are a few reasons I keep my CC subscription and one of them is that Photoshop is one of only two applications I know of that gives me the maximum control possible over prints I make on my inkjet —- and my fear is that they will throw away in a fit of FOMO.
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.
There are more than dozen no-cost, shared-code projects creating model harnesses for software creation and already some of the no-cost, self-hostable models are OK in those harnesses. Not as good as the subscription based, hosted harnesses and models, but OK.
If we assume that the no-cost harnesses, models, and cheapish self-hosting rigs improve over time then at some point they'll be good enough that they serve almost everyone who needs it.
Then only fools will pay for software and only services like support will remain in business... oh except that no-cost chatbots will fill those roles, too.
Then all that's left is mathematicians... oops.
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.
The point here is a different one I think. It's not that everyone and their dog will create their own version of things, this will probably not happen at such a huge scale for the reasons you mentioned.
What might change is that the barrier to enter the market is lowered so someone without gigantic investment and years of time could actually come out with e.g. a JIRA competitor ticking all the boxes. This drives the vendors margins down. Also might give the buyer more negotiation power if you can say "then we just do our own version".
There is a lot of vendor lock in the software world just because it's just too hard/risky to do a full featured competitor, that might change and eat into the margins of especially those vendors. For those who also have healthy competition, the relative dynamics probably don't change much, agreed.
It’s hard to predict a future, there are so many ways it could play out.
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”
Just like railroads and telecoms back in the day. Even if they go under, the infrastracture built will benefit the next generation who will find a good use for it
There is a difference though. Railway lines and fiber last decades, GPUs will be out of date in a few years and the chips will probably just die at some point anyways. Also datacenters are barely getting finished and GPU clusters sit in storage. They are kind of useless without those datacenters though as they need specialized infrastructure.
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.
Large AI firms are hostile to their customers. Given their surveillance, censorship, blocks, deprecations, and bugs, it doesn't make sense for users to stick with them for a day longer than necessary.
The article is wrong. Adobe charged rent to create profit to send to shareholders. Paying engineers was not the major cost center other than that it needed to _seem_ like it was, to justify the subscription price.
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.
I hate that this article is talking about Photocraft killing Photoshop because of AI as though Ivan Kutskir didn't build Photopea years ago by hand. It was an amazing feat and Photopea is a great product and I don't want to see it dismissed immediately just cuz someone copied it over a weekend with Claude.
Photopea has ad's though- Photocraft doesn't- he may of made Photopea "by hand" but it is a direct copy of Photoshop- "AI" copies better and faster than "Ivan Kutskir " and doesn't include ad's- dude was just a temp human thief that needs to sleep and eat (hence the ad's) AI is a far superior thief
For the record, I believe that the whole "building something as good as Photoshop" is BS, and a sort of retroactive justification of lock-in.
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."
Outside of specialized niches, toll booths exist due to network effects.
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.
Word is better than the alternatives. In Germany they keep going through the cycle of open source zealots mandating that government workers use off-brand office suites and then their unions going on strike to get Word back.
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.
I disagree with your use of the word zealot but I agree, MS Office is objectively the best office suite. I no longer use it daily, but anybody arguing otherwise has: never used it, never tried the alternatives to see how bad it can be, doesn't do collaborative work, or doesn't need any features beyond bold/italic. For example I can do most of my coding in Notepad, but it's still a very inferior experience. I feel the same way about MS Office vs LibreOffice.
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.
If there is anything else the world is learning, it’s that open source developers don’t have the same priorities as eneterprise or governments need for successful software use at scale.
I'm not talking about open source, I'm talking about the risk for non-US governments to be dependent on a sole supplier to be able to function, be it Microsoft or Starlink or any US-based corporation.
Greed doesn't let people to see far enough. Investors of AI see it as a once-in-a-century kind of tech revolution and they don't want to miss riding it. Fund managers will be fired if they show any skepticism of AI. It's the myopic and FOMO forces that are at play. Banks would jump into the great Tulip mania if it happens again without blinking an eye. JP Morgan's CEO who once called crypto as idiotic, went full steam and joined the "idiotic" bandwagon. Not because he got enlightened, but because of FOMO.
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.
While I want to agree with much of this, the foundational flaw is that the Ai is not as capable as though author wants us to believe. Perhaps they are being intentionally hyperbolic about it.
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.
The company I work for has had 2 or 3 rounds of layoffs over the past year and a half. CEO and VPs haven't got laid off or a paycut. I think cuts should start with leadership. They need to own the mistakes that were made.
So many times you see reporting on a round of layoffs and it quotes the CEO saying “I take responsibility.” But… how? It seems like the actual responsibility is being borne by the people who just lost their livelihood. What exactly is this definition of responsibility where you experience no negative effects whatsoever?
Their lot (VP-level certainly) rotates in and out, often with 3-4 layers of middle management between them and the crews executing, which makes direct assignment of blame exceedingly difficult. Every 3-4 years you get a near-complete rotation of leadership in an area of †he business, and any memory of what anyone was trying to achieve before is forgotten, either intentionally or unintentionally, and then you do it all over again.
I don't think anyone killed old school engineering, whatever that means for you. But if someone indeed did kill it, incompetence, management and incompetent management is the culprit
I'm surprised so many people not seeing it. The Claude voice is all over that article, full of "It's not X, it's Y" and other signs. Half the points don't even make logical sense.
The Chinese companies will not stay "open" if it doesn't serve their economic/political interests.
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?
Interesting point, overlooking the biggest part:
"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.
> Massive toll
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..
> 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.
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.
Knowledge and labour don't magically exclude the AI labs. Who still needs to get paid in such an endgame?
dead labor something something
Oh but why do they need a high salary when AI gives us free software? /s
Models only have information up to the date when they were generated. APIs get updated and expanded, new frameworks and libraries get introduced, bugs get fixed. How long is an existing model going to still be useful? A year or two at most?
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'm using Claude on APIs on which it was not trained, and it's doing well. The techniques it uses to debug and find solutions aren't that different from older APIs. "Let me dump all of this response and guess which field might be what" sometimes, "let me get a stack trace just before it goes wrong", many familiar development techniques.
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.
> APIs get updated and expanded, new frameworks and libraries get introduced, bugs get fixed. How long is an existing model going to still be useful?
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.
These companies are burnings billions each year to train new models for .. nothing ? Out of pure stupidness ? There probably is a reason
Yes there are reasons to create better models. Baking in latest revision of assorted API documentation is not it.
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.
This reminds me of a funny experience at work
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;
Not even human's can keep track with all the latest APIs.
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.
Opus 5.5 can write IBM 1401 Autocoder very well (used on the Computer History Museum's pair of working 1401s). THAT is astounding to me.
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.
Case in point, MiMo-2.6 was RL'd for $3.5M
https://mimo.xiaomi.com/rl/
Models can ingest up to date documentation, source code and search the web, context windows are big enough these days. Sure at some point retraining will be needed but not every few month and you likely don't have to start at zero every time.
https://en.wikipedia.org/wiki/Retrieval-augmented_generation
RAG is not the answer
Docs, OpenAPI schemas, web fetch, and scripts to understand the request/response are used by models today. It works for any API over time.
Why do we accept this "massive cost" phase all the time? I want to see the data and intelligent analysis of it. I think this is not clear yet.
Current models do well with proprietary systems they were never trained on. Ar most you need a good AGENTS.md and a directory full of skills.
Red herring. AIs can read an api spec and do just fine.
Your last paragraph doesn’t hold water.
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.
the font on that site is sooo hard to read
There are no "free models", you know ?
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;
[flagged]
Please expound.
The irony here is that the article, written solely by Claude, is also a volley at authors.
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?
Does anyone actually use PhotoCraft? Is it really a complete replacement for Photoshop or is it basically a proof of concept?
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.
I tried the Lightroom alternative, it was garbage, not even close to a replacement. They have spent decades fine tuning the smallest of features, you’re not going to replace that overnight.
I’m not a hardcore Lightroom/Photoshop user, but I have hundreds of RAW product photos that I need to cycle through a few times a year, and a few new ones to add whenever I make a new product. I long-since canceled my CS subscription because it wasn’t worth hundreds of dollars a year to me.
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 use it everyday- I don't need Photoshop anymore- its faster and does everything I need it to do- the only other "creative" artcraft app I've really tried to use is Filmcraft, the Premiere replacement but I'm still using Davinci Resolve Studio as Filmcraft isn't as far along as Photocraft but Davinci Resolve Studio is slow, bloated, can't customize the UI and barely has an API so I can't wait to dump it
It's not an ideal replacement as of yet, but the writing is in the wall and I bet in a few year time it (or some other clone) will be quite mature and stable.
The writing isn't on the wall until Artcraft is
1) As stable as Photoshop.
2) Competitive against Photoshop among professionals.
Until it hits the first point, which it hasn't yet, it's just slop. AI enjoyers don't care how broken it is and will use it anyway just because it's AI generated but everyone else will stay away, especially most pros.
If it can't hit the second point then at best it's just another image editing app, and those already exist. Adobe isn't afraid of Krita or Photopea.
I'm using it daily. It's not as good as Photoshop with brushes, the quick selection and masking features and the type engine is not as sophisticated. But for the simple edits I need to do, it is more than adequate.
Earlier this year we had prospective customers do comparisons against our competitors but also against prompting their own. All of them, without exception, found that to be a bad idea.
If it were just about the ability to edit images, Gimp would have obsoleted Photoshop decades back. Photoshop excels because of its capabilities but even more because of its ease of use and the stickiness of the skillset its users have spent developing.
Nobody uses it. It’s completely broken and was dead on arrival. It can’t even reliably save files. Maybe eventually it’ll get to a sort of working state, but I doubt 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.
In short, a demonstration of commitment.
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.
On the other hand, we are barely a month in, that thing is still ongoing so it's hard to judge now.
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?
I'm watching Photocraft because if 1000 people on the Internet working together can vibe code a viable Photoshop clone we will truly be in a new era.
> that money doesn’t migrate to the AI lab. It evaporates.
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!
> No, someone still has to pay the subscription. AI Labs don't care who uses the tokens and how.
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.
That's the point. Adobe misses out, not the AI companies.
And why would the AI lab care what happens to adobe?
Because supposedly Adobe would have paid a lot more for AI and tokens than a couple of random people working on Artcraft.
No one has to pay for tokens with open weight models.
No one will be able to pay for tokens if their product is replaced by OSS created by open weight models.
> AI Labs don't care who uses the tokens and how.
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.
> 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.
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.
> open models
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
just an aside on the Claude prompter responsible for this article, he labels himself as writing about how "feminism ruins tech", and he has written some pretty heinous things about women. Given that he approaches things with such a (male) software engineer exceptionalist point of view, I would take his writings with a grain of salt.
[flagged]
There are a few reasons I keep my CC subscription and one of them is that Photoshop is one of only two applications I know of that gives me the maximum control possible over prints I make on my inkjet —- and my fear is that they will throw away in a fit of FOMO.
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.
There are more than dozen no-cost, shared-code projects creating model harnesses for software creation and already some of the no-cost, self-hostable models are OK in those harnesses. Not as good as the subscription based, hosted harnesses and models, but OK. If we assume that the no-cost harnesses, models, and cheapish self-hosting rigs improve over time then at some point they'll be good enough that they serve almost everyone who needs it. Then only fools will pay for software and only services like support will remain in business... oh except that no-cost chatbots will fill those roles, too. Then all that's left is mathematicians... oops.
Long time SRE/DevOps person here:
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.
The point here is a different one I think. It's not that everyone and their dog will create their own version of things, this will probably not happen at such a huge scale for the reasons you mentioned.
What might change is that the barrier to enter the market is lowered so someone without gigantic investment and years of time could actually come out with e.g. a JIRA competitor ticking all the boxes. This drives the vendors margins down. Also might give the buyer more negotiation power if you can say "then we just do our own version".
There is a lot of vendor lock in the software world just because it's just too hard/risky to do a full featured competitor, that might change and eat into the margins of especially those vendors. For those who also have healthy competition, the relative dynamics probably don't change much, agreed.
It’s hard to predict a future, there are so many ways it could play out.
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”
Just like railroads and telecoms back in the day. Even if they go under, the infrastracture built will benefit the next generation who will find a good use for it
There is a difference though. Railway lines and fiber last decades, GPUs will be out of date in a few years and the chips will probably just die at some point anyways. Also datacenters are barely getting finished and GPU clusters sit in storage. They are kind of useless without those datacenters though as they need specialized infrastructure.
I generally agree with the diagnosis here, but:
> 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.
Is this not simply the tendency of the rate of profit to fall, just a bit extreme?
Large AI firms are hostile to their customers. Given their surveillance, censorship, blocks, deprecations, and bugs, it doesn't make sense for users to stick with them for a day longer than necessary.
Would be fascinating to be a fly on the wall at the Adobe offices currently
The article is wrong. Adobe charged rent to create profit to send to shareholders. Paying engineers was not the major cost center other than that it needed to _seem_ like it was, to justify the subscription price.
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.
I hate that this article is talking about Photocraft killing Photoshop because of AI as though Ivan Kutskir didn't build Photopea years ago by hand. It was an amazing feat and Photopea is a great product and I don't want to see it dismissed immediately just cuz someone copied it over a weekend with Claude.
Photopea has ad's though- Photocraft doesn't- he may of made Photopea "by hand" but it is a direct copy of Photoshop- "AI" copies better and faster than "Ivan Kutskir " and doesn't include ad's- dude was just a temp human thief that needs to sleep and eat (hence the ad's) AI is a far superior thief
For the record, I believe that the whole "building something as good as Photoshop" is BS, and a sort of retroactive justification of lock-in.
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."
Outside of specialized niches, toll booths exist due to network effects.
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.
Word is better than the alternatives. In Germany they keep going through the cycle of open source zealots mandating that government workers use off-brand office suites and then their unions going on strike to get Word back.
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.
I disagree with your use of the word zealot but I agree, MS Office is objectively the best office suite. I no longer use it daily, but anybody arguing otherwise has: never used it, never tried the alternatives to see how bad it can be, doesn't do collaborative work, or doesn't need any features beyond bold/italic. For example I can do most of my coding in Notepad, but it's still a very inferior experience. I feel the same way about MS Office vs LibreOffice.
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.
Objectively?
I use Libreoffice and I ... objectively... find it slightly less of a pain in the ass.
> cycle of open source zealots
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.
If there is anything else the world is learning, it’s that open source developers don’t have the same priorities as eneterprise or governments need for successful software use at scale.
20 years ago called and wants its take back.
Open source developers work on Linux. What you're saying presently makes no sense at all.
I'm not talking about open source, I'm talking about the risk for non-US governments to be dependent on a sole supplier to be able to function, be it Microsoft or Starlink or any US-based corporation.
I still have a copy of Photoshop 7 burned on a CD. Photoshop was once really excellent, back before its glory was clouded by subscriptions.
Let the time be the judge!
Why wait when I can use LLM as a judge
Greed doesn't let people to see far enough. Investors of AI see it as a once-in-a-century kind of tech revolution and they don't want to miss riding it. Fund managers will be fired if they show any skepticism of AI. It's the myopic and FOMO forces that are at play. Banks would jump into the great Tulip mania if it happens again without blinking an eye. JP Morgan's CEO who once called crypto as idiotic, went full steam and joined the "idiotic" bandwagon. Not because he got enlightened, but because of FOMO.
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.
While I want to agree with much of this, the foundational flaw is that the Ai is not as capable as though author wants us to believe. Perhaps they are being intentionally hyperbolic about it.
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.
"Software engineers are the most expensive employees on Earth"
They are not. CEOs are.
The company I work for has had 2 or 3 rounds of layoffs over the past year and a half. CEO and VPs haven't got laid off or a paycut. I think cuts should start with leadership. They need to own the mistakes that were made.
So many times you see reporting on a round of layoffs and it quotes the CEO saying “I take responsibility.” But… how? It seems like the actual responsibility is being borne by the people who just lost their livelihood. What exactly is this definition of responsibility where you experience no negative effects whatsoever?
Their lot (VP-level certainly) rotates in and out, often with 3-4 layers of middle management between them and the crews executing, which makes direct assignment of blame exceedingly difficult. Every 3-4 years you get a near-complete rotation of leadership in an area of †he business, and any memory of what anyone was trying to achieve before is forgotten, either intentionally or unintentionally, and then you do it all over again.
> I think cuts should start with leadership.
That would be fair, but leadership decides who to cut, so... here's your belongings, please be out of the building by noon.
Yes but a company only needs 1 CEO while they are paying for 100x or 1000x of SDEs
And that one CEO is paid 100x to 1000x more than their SDEs, while SDEs create more value for the company.
But, you are moving the goal post.
Your quote is misleading and makes it sound like the author made this point.
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.
[dead]
Claude is everywhere. It literally killed oldschool engineering. I don't even want to interact with people who use Claude anymore.
I don't think anyone killed old school engineering, whatever that means for you. But if someone indeed did kill it, incompetence, management and incompetent management is the culprit
If only management was to blame, we wouldn't see the AI spam in the open-source world.
The fact that we do proves that ICs are equally to blame, because evidently the majority of them love using AI.
> The fact that we do proves that ICs are equally to blame, because evidently the majority of them love using AI.
Quite true. Although there is a distinction, you definitely can use it as a tool but many indeed just lazily create slop.
Good.
Props to the author, this is very well written. Makes a lot of sense.
The author here is Claude (possibly something that distilled it).
Dude was like “write me a diatribe about AI arguing why coding agents don’t make sense”
I'm surprised so many people not seeing it. The Claude voice is all over that article, full of "It's not X, it's Y" and other signs. Half the points don't even make logical sense.
I’ve become hypersensitive to “Claude voice” and still can’t persuade myself that this was just a Claude prompt.
I'm happy my NextDNS filters blocked the domain for me, saved me a click
Yea, the article is littered with Claudisms. You can see the: "it's not this, it's that."
The Chinese companies will not stay "open" if it doesn't serve their economic/political interests.
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?