Over 100 Public Software Companies Getting 'Squeezed' by AI, Study Finds (businessinsider.com)
- Reference: 0177076131
- News link: https://tech.slashdot.org/story/25/04/21/1443240/over-100-public-software-companies-getting-squeezed-by-ai-study-finds
- Source link: https://www.businessinsider.com/software-companies-squeezed-by-ai-alixpartners-2025-4
The squeeze reflects a dramatic shift: AI agents are evolving from mere assistants to becoming applications themselves, potentially rendering traditional SaaS architecture obsolete. High-growth companies in this sector plummeted from 57% in 2023 to 39% in 2024, with further decline expected. Customer stickiness is also deteriorating, with median net dollar retention falling from 120% in 2021 to 108% in Q3 2024.
[1] https://www.businessinsider.com/software-companies-squeezed-by-ai-alixpartners-2025-4
Innovate or Die (Score:2, Informative)
This is not new. Not sure AI is really the right direction yet, but in tech, innovation has always been the key to long-term success. Whether that innovation is acquired or developed in-house, well, that's a different question.
Black-boxification of software (Score:5, Insightful)
Isn't this going to make software even more impenetrable than it already is? Traditional software - even that with lots of tech debt and shitty or non-existent documentation - is amenable to analysis, troubleshooting, and being fixed. Its behaviour is largely deterministic, is it not?
Is the same true of 'applications' which are mostly queries to an LLM? If I query an LLM today, and issue the same query to the same LLM in a week, will the answer be identical? If not, then it seems to me there's potentially a hell of a problem with "AI agents (are) evolving from mere assistants to becoming applications themselves".
I have almost zero expertise here, so please tell me what I'm missing in this picture and if I'm right, wrong, or somewhere in between.
Re: (Score:2)
For LLM codegen, the problem really is the same- trying to figure out someone "else's" code.
In general, I can say that LLM codegen'd code is usually far better documented, though. If you ask it to crank up the verbosity, it'll happily do so.
They tend to produce pretty clean code, but definitely not without mistakes.
In general, I haven't seen where the "prompt' used to create it is saved, because frankly... it's just not that simple.
The "prompt" in the terms of an LLM is really its entire context window,
Re: (Score:2)
> In general, I can say that LLM codegen'd code is usually far better documented, though. If you ask it to crank up the verbosity, it'll happily do so. They tend to produce pretty clean code, but definitely not without mistakes. In general, I haven't seen where the "prompt' used to create it is saved, because frankly... it's just not that simple.
Thanks - I suspected that last part about the prompt being saved, and it's good to have confirmation. Regarding documentation, if you as the programmer fix a problem in the code that the LLM generates, is there a mechanism to track the change and any comments about it? And is the LLM in any way "aware" of the changes and comments?
Do you think things like comments, versions and change tracking will be introduced to make these tools more robust and manageable? Or are we at the point where business cases and f
Re: (Score:2)
Get a local model (or an API provider that guarantees you not to change the model), fix the parameters (including all random seeds) and you get deterministic behavior. The question is, if you're always sure that the first shot is the best. And if the input you process is the same.
Re: (Score:2)
Thanks for the info. For me, this also raises the question of what happens during a software update. Are bugs or incorrect answers more likely to occur when updating the LLM and/or its 'app' than when updating a compiled program of similar function and complexity?
what are the companies? (Score:3)
What are the companies and what do they do. Article is behind a paywall. Summary is a bunch of AI buzzwords. No information here.....
Re: (Score:2)
Thanks, but this is a news aggregator site. Doing these simple things is the entire point of it.
Re: (Score:2)
Actually, the article also doesn't have the list of companies, or even a link to the report. Can you track that down for me?
Re: (Score:2)
Mom, slashdot is linking to the original article again!
Re: (Score:2)
Take a look at the "Tranquility Reader" extension for Chrome / Firefox, it can display some of those paywalled articles, such as this one.
Re: (Score:2)
On Firefox without add-ons, current method is to enter "Reader Mode" which is the first icon to the right of the URL. The traditional method (still valid, but ugly) is to use the good old menu bar (still available from the right click) then Display/Page Style/None (which deactivates CSS).
Re: (Score:2)
What paywall? I read the entire article without issue. The problem must be on your end.
That said, they don't specifically mention any company. Only the general term of mid-sized SaaS companies.
Bake or Fake (Score:2)
How does the public know whether this is a real phenomena or companies merely blaming their slack or recession on AI because they can?
Bots still make way too many dumb mistakes.
Correction (Score:1)
Grammar correction: should be "phenomenon" and not "phenomena".
Re: (Score:3)
I suspect that at least some of it is unrelated.
One of the reasons why so many vendors embraced 'AI' so desperately is that, just before the 'boom' really took off, things were not looking so good in 'SaaS' world. VCs and stockholders wanted infinite growth forever, obviously, but in 2023-ish, if I'm remembering correctly, average number of SaaS products in service per enterprise declined after something like a decade of growth; and individual companies in the area were seeing other metrics(customer acqu
Re: (Score:2)
> How does the public know whether this is a real phenomena or companies merely blaming their slack or recession on AI because they can?
> Bots still make way too many dumb mistakes.
I'l let you in on a little secret, the companies don't give a fuck about the results as long as they can save money. Let's say they buy a customer service bot from Googs and it is utterly worthless. Why does that matter to the company when they already have your money? You can't go to the government anymore for help as mango mussolini and his dorks systematically dismantled anything that would help you. It's a rigged system, a scam. Funnel all the money to the already wealthy and you get left with eith
Spaghetti AI = consultant boom (Score:2)
> companies don't give a fuck about the results as long as they can save money. Let's say they buy a customer service bot from Googs and it is utterly worthless.
One way to interpret this is similar to the offshore-to-India fad where companies did indeed get software made relatively cheaply, but the maintenance costs kept skyrocketing, as duck-wire needed chicken-tape to get it to work. Such companies eventually brought those projects back inhouse, and consultants made boat-loads of money fixing the offs
Re: (Score:2)
For the record, I'm not saying E. Indians are "bad coders", only that the coding shops were rewarded for delivering initial features, not for making the software maintenance-friendly. So the offshorers got what they paid for and only what they paid for. The bosses didn't understand the importance of long-term maintenance so didn't measure/reward for it.
Re: (Score:2)
> The bosses didn't understand the importance of long-term maintenance so didn't measure/reward for it.
Or they understood it wasn't worth the cost. Corporations have a single value, the bottom line, and measure decisions against that value. There are certainly employees who have other values but applying those values in ways that hurt the bottom line are a violation of their fiduciary responsibility to the company.
Re: (Score:2)
Fiduciary responsibility does not, and has never meant, that the bottom line must continue to go up.
I have to imagine that anyone who believes that dumbass meme has never, never worked in a corporation where they were actually exposed to the C suite.