A lot of the AI industry is powered by VC money at the moment. Which means a lot of the services offered are using very low fixed billing. This loses a ton of money, but is seen as an investment. The idea is to get a lot of customers and lock them into the service. Then slowly raise prices, enshittify the service and thus gain back the money lost in the beginning.
However what we’ve seen in this AI bubble is everything is moving too fast for that to be a good option. VC money is running out fast and the price of AI from the large providers is getting higher by the day. This means a lot of those AI startups that were trying to get customers are losing money much faster than expected and either have to raise prices or shutdown.
Especially in sectors where there isn’t a lot of money to be made. Customer support for example is one of those areas that costs money and doesn’t make money normally. So corners get cut, people don’t get paid much and stuff is outsourced to India for example. When the AI solutions are offered at the low fixed price, it’s a good idea for companies to use it. But when the price goes up and pricing becomes unpredictable if token based billing is involved, suddenly it isn’t as good an alternative as the older solutions were.
This is a large part of the AI bubble when people talk about that. It isn’t just the circular investments, way too high spending commitments and overpromising of the tech. It’s also the huge amount of VC based startups and investment programs that can’t economically exist without that continuous injection of funds. Now that the real price of AI is showing up, it looks like a lot of those startups will go bankrupt and large projects funded by investment will stop. With a lot of the market price of the larger tech companies being based on this huge demand that might evaporate in short order, the fear is a lot of them are way overvalued. When the market corrects for this, the impact will be pretty big and the fallout felt by everyone.
Thank you.
This is exactly it.
The VCs are following a well worn script - the freemium model worked well for SaaS.
The problem is that the cost of customer acquisition in SaaS is high relative to the long term profitability of the customer (it doesn’t cost too much to add capacity for another M365 user), while for AI it’s the inverse - the more the customer uses the product, the more cost they generate for the AI provider. Unless that cost can be returned to the customer, in the form of per-token billing, the math doesn’t work out to a positive income.
So not only do you have a sharp increase in cost when the company stops burning VC money and needs actual revenue from customers, but the increase goes from a flat rate to a pay as you go. GitHub Copilot made a change like that earlier this year and caused havoc, but i think that the token rate is still lower than cost.
Thank you.
So if I spent hours taking bollocks to an ai assistant, it would burn up money for either the ai company or the company using AI? If I’ve got nothing better to do, is there a downside?
Downsides include environmental impact, and the usage might skew the company’s involved to belief it’s more useful than it really is
From personal anecdotal experience: chatting for hours would cost the company a few cents at most. In our company, developers are encouraged to try it out, and there are a few projects that use it a lot. The user with the highest cost has been using it in a loop, with minimal human intervention (and hence downtime), and it cost the company 2000 bucks for a month.
With “normal” use, even extensive, it’s very hard to go above 50 bucks.
I don’t know the company’s agreement with the provider - I’m guessing this is on top of a subscription cost - but either way if you want to make a real dent just chatting isn’t going to do it.
Ask it to calculate a few billions of places of pi between each answer
Nah, word volume between you and it burns tokens the most, so post the full bee movie script, and ask it to write a 5k essay about it, over and over
specifically, output tokens. The more it has to write back, the more it costs
Things is… those customer AI agents aren’t better than what we Had before. They dunno jack Shit about it to actually make the expense reasonable. Also fuck AI.
Are you ok with a chess bot cuz that’s technically AI?
So is an automatic transmission. AI in this context is specifically referring to LLM/generative AI.
Wouldn’t that be programmed? Then it’s not AI.
Technically it still falls under the overall AI umbrella but not the LLM/GenAI slop subset which has taken over the label.
Depends on what contract the company has made with the AI company.
I suppose so. I understood how SAAS and cloud stuff work with the billing.
But I don’t understand how with the token nature of AI lifespan relates to the AI customer service agents.
Do you have any examples of how an standard contract like that be structured ?
Mostly it’ll be per token based billing. How h depends on the quality of the model being used.
Companies will often use a platform like AWS Bedrock or OpenRouter and use models through that. It gives them a consistent interface to many different models from many different providers under a single contract.
In most cases the client company is too small for the AI company to be worth making a custom contract, so just look up the standard contacts from some of the big AI providers.
