'Software Isn't Dead, But Its Cosy Business Model Might Be' (ft.com) 27
The software industry's decades-old habit of charging companies a flat fee for every employee who uses a product is running into a fundamental problem: AI agents don't sit in chairs, and they don't need licences.
As autonomous agents take on tasks that human workers once handled, the per-seat pricing model that made SaaS revenue so predictable is giving way to consumption-based and hybrid alternatives. Snowflake and Databricks (valued at $134 billion) already charge based on usage. Salesforce initially priced its Agentforce customer relations bot at $2 per conversation but faced customer pushback and now offers action-based pricing, upfront credits and fixed fees.
ServiceNow's finance chief Amit Zavery said last month that some customers aren't ready for purely consumption-based models. Goldman Sachs estimates US software spending will nearly triple to $2.8 trillion by 2037 as automated tasks blur the boundary between IT and wage budgets, but that money will no longer arrive in the neat recurring instalments that investors and private equity firms have come to expect.
As autonomous agents take on tasks that human workers once handled, the per-seat pricing model that made SaaS revenue so predictable is giving way to consumption-based and hybrid alternatives. Snowflake and Databricks (valued at $134 billion) already charge based on usage. Salesforce initially priced its Agentforce customer relations bot at $2 per conversation but faced customer pushback and now offers action-based pricing, upfront credits and fixed fees.
ServiceNow's finance chief Amit Zavery said last month that some customers aren't ready for purely consumption-based models. Goldman Sachs estimates US software spending will nearly triple to $2.8 trillion by 2037 as automated tasks blur the boundary between IT and wage budgets, but that money will no longer arrive in the neat recurring instalments that investors and private equity firms have come to expect.
Slashdot is what is dead (Score:1)
Re: (Score:3)
The fact that they were charging $2/conversation is interesting. My Eliza Bot could very slowly bankrupt them.
Re: (Score:2)
$2 per conversation is INSANE.
I hope these are the most insightful, life changing conversation.
Cause I used the chat bot of a subscription service yesterday and it's so bad I decided to call and wait 2 hours on the phone instead.
Re: (Score:2)
Not really insane. If the AI can handle conversations with customers that are being done by a person before, and the person can handle 10/hour on average (6 minutes per conversation), then $20/hour for the AI is a bargain (assuming the same time spent on each call).
Also assuming the AI is way better than a chat bot.
Cheaper than the persons wages, and a lot less than the total cost of the employee (which is typically a multiple of their wages).
Re: (Score:3)
Clearly the paywalled article business model is dead.
Lol, what? (Score:5, Informative)
>AI agents don't sit in chairs, and they don't need licences.
Yeah, right. I don't think so. If it is using the software, it is going to need a license. Multiple agents, multiple licenses. It's not hard.
Re: Lol, what? (Score:2)
Re:Lol, what? (Score:4, Insightful)
In many cases it's "multiple copies running" and that's been a thing for decades. Each agent is going to need another copy running any time it's using the software. If you want to have all the agents queue up all of their requests and then have a single worker process serve all of them then you might reduce the number of licenses used, but it won't necessarily reduce them to 1 depending on the load. And this sort of thing has already long been done, with workers in specific rules doing all of a certain type of processing for example.
Re: (Score:2)
Re: (Score:3)
You're confusing science fiction with reality again.
Re:Lol, what? (Score:4, Funny)
Re: (Score:2)
Re: (Score:2)
Yeah, if that were possible, maybe, but that's not reality. AI will get you a bloated and buggy "first draft" that is impossible to maintain very quickly, but you'll soon discover that trying to wrangle that mess is far more expensive and time-consuming that just hiring competent people to write and maintain the software properly.
Nonsense article like this one always assume that AI is actually as capable as Sam Altman promised us years ago, despite all of the evidence to the contrary. We might as well be
Re: (Score:3)
Read past the first paragraph of the fine summary.
They are not proposing "FREE for bots". They are proposing that rather than a flat rate charge per Seat license, they charge by Time of Use or per Query. Each bot doesn't need a license, but the company does -and it gets billed for usage.
False premise... (Score:5, Insightful)
I wouldn't be surprised if the juiciest days of SaaS rent seeking are behind us(if nothing else, SaaS vendor numbers were starting to look less promising prior to the 'AI' craze; arguably one of the reasons why they all jumped on it like rabid animals hoping that it would salvage their growth); but this premise seems deeply and obviously flawed. Per-seat licensing has never involved chairs; and (especially when you are dealing with software contracts high value enough that you can litigate, rather than relying purely on DRM) you can make whatever you want need a license.
There's obviously no completely ironclad way to stop your customers from using a scraper to hide their activities; just as you can't entirely prevent account sharing between employees who should be licensed separately; but there's nothing about 'agents' that is any harder to require a license for.
Who wrote this article? (Score:2)
A CEO who's woken up to the realisation that all this useless AI costs heaps of money for even the simplest of tasks?
Sounds like the only reason why this article was written was to cosy up the public to pay more soon.
The AI investment pond must be drying up very quickly out there, with sheer desperation for any returns.
What tasks? (Score:2)
I want one article that states clearly and unambiguously what tasks Ai is replacing human workers on.
Re: (Score:3)
I want one article that states clearly and unambiguously what tasks Ai is replacing human workers on.
Writing articles, it seems - it's pretty obvious from the output.
Re: (Score:2)
Touche
My heart bleeds for SaaS (Score:2)
And the world's smallest violin, played by a tardigrave, is the music.
SaaS was *always* a terrible idea for anyone but the rent-seekers.
Re: (Score:2)
SaaS was *always* a terrible idea for anyone but the rent-seekers.
Only a sith deals in absolutes.
Now, don't get me wrong, as a rule, I prefer CapEx license purchases; there *should* be software that just-runs once it's feature complete.
However, I can think of three instances of SaaS that predate what we know as SaaS today: Delorme Street Atlas, Norton Antivirus (or AV in general), and TaxCut/Turbotax. In each of these cases, there are legitimate content changes that justify continued work, which requires legitimate development time. Streets get added, removed, and changed
Re: (Score:3)
However, I can think of three instances of SaaS that predate what we know as SaaS today: Delorme Street Atlas, Norton Antivirus (or AV in general), and TaxCut/Turbotax. In each of these cases, there are legitimate content changes that justify continued work, which requires legitimate development time.
I would argue that for Street Atlas, the software isn't changing; the data is. Getting map updates should cost money. Continuing to use the software with the existing map data shouldn't. So SaaS is still bad. DaaS (data as a service) is less so.
For antivirus, the same thing is true.
And in both cases, limiting the software to a single data provider harms consumers by forcing a software change if you change data providers. In an ideal world, other map providers would be allowed to write data files compat
Seems to me this means finding new ways to charge. (Score:2)
Seems this means to find more ways to extract money from customers. With backup programs, it was servers and clients. Now, it is capacity, and the costs of backing up a typical NAS can be insane, when in a previous version, the NAS was just another client.
The problem is overcharging and making every iteration worse for customers, assuming they will stay with you... or your software is a Veblin good.
In the past, things happen, then a new software company comes in and sweeps the old guys aside. Veeam came
needlessly ambiguous (Score:2)
I don't have a big issue with consumption based pricing in principle. But I *do* have a problem with the needlessly ambiguous units of measurement which feel purposely enshittified to confuse customers.
if I'm looking at switching products/services I really need to understand cost. I don't want to spend 2 hours figuring out what a DPU or a credit or a token actually is. For reference snowflake's pricing/utilisation info is here [snowflake.com]. It's 19 pages long.