Hooked.
Most of us didn't decide to rely on AI. It just crept in.
A draft email here. A meeting summary there. A spreadsheet formula you'd never have worked out on your own. A year ago it was a novelty. Now, for a lot of people, ChatGPT, Copilot or Claude is open in a tab all day, the way a calculator used to sit on the desk.
So it's a fair question: what happens when it stops being cheap? Most of these tools are free or cost about the same as a streaming subscription. That feels too good to last. Surely, at some point, the bill arrives.
This week we got an unusually honest look at the question. Anthropic, the company behind Claude, published the document it has to file before selling shares to the public. Companies rarely say this much about themselves, and it's written for people deciding whether to hand over their money, so it has to list what could go wrong.
I read it with one question in mind: should ordinary users and small businesses be bracing for a price shock?
My answer is no. But there's something else you should be bracing for instead.
Why the price probably isn't the thing to fear
The headlines last week were about an enormous loss. Most of that figure, by the company's own account, is an accounting entry linked to how it borrowed money and how much more valuable it has become. It isn't money spent answering your questions.
What matters more for anyone using these tools is the direction of travel, and in the filing it's down. Anthropic says that when it released a newer model in September, it cut the cost of typical work by about 25 per cent compared with the one before. Newer models tend to do the same job with less effort, and the companies compete hard on price.
There's a simple reason for that competition. These companies want you in the habit. The filing is frank that free access costs real money to provide, and that a big part of the point is to build familiarity with the product. That's the business model working as intended, not a charity about to run out of funds.
I should say the obvious. I have a horse in this race. M-Tech builds things for clients on these tools, and back in June I wrote that the price barrier for small businesses had fallen away. So treat this with the scepticism it deserves.
The fair objection
The best argument on the other side is a good one, and it goes like this.
Building the next generation of these tools costs a fortune. The companies are spending vast sums on computing power and researchers, and soon they'll have shareholders expecting a return. Free plans could shrink. Cheap plans could get less generous. Something you use daily could quietly move to a higher tier.
I think some of that will happen. Free plans in particular are a promotion, and promotions change.
But that's the normal risk of any supplier, the same as your phone contract or your accounting software. It's a reason to keep an eye on things, not to panic. Several big companies are competing for the same customers, and that tends to keep prices honest.
What actually changes under your feet
Here's the risk I'd take more seriously, because it's quieter and almost nobody plans for it.
The price stays the same, and the thing you're paying for changes.
The filing says it plainly: "We have also changed, and may in the future change, usage limits, rate limits, or model and product capabilities." It also says older models get retired, and that people who have built them into their work may resist moving to the new ones.
In everyday terms, that looks like this. The assistant that used to handle your long documents now tells you you've hit your limit by lunchtime. The version that wrote in a way you liked is replaced by one that's cleverer but different, and the template you rely on stops coming out right. A feature you'd built a weekly routine around moves, or goes.
Think of it like a bus route. The fare is predictable enough. What catches people out is the route changing, and nobody telling them until they're standing at the wrong stop.
We've already seen the extreme version. In June a newly launched Claude model was switched off for everyone for nearly three weeks after the US government stepped in. I wrote about it at the time in Vanished. and Back.. The filing now tells investors, in writing, that something similar could happen again.
None of this means the companies are behaving badly. Making these tools better means changing them. But it does mean that "it worked yesterday" is a weaker promise than it is with most software.
What I'd do this week
Nothing expensive, and nothing that needs a consultant.
- Notice what you actually lean on. Not "I use ChatGPT", but which jobs you'd struggle to do this week if it vanished. For most people it's two or three tasks. Write them down.
- Keep your own copy of anything that matters. If you've crafted a set of instructions that gets a report or a letter just right, save them somewhere that isn't inside the tool. If the tool changes, you start again from your notes rather than from memory.
- Try a second tool once. Run one of those everyday tasks through a different assistant. Not to switch, just so that if your usual one changes or goes down, you're not learning from scratch on a deadline.
- If you run a team, ask the same question at work. Which processes now quietly depend on one AI tool behaving exactly as it did last month? If the answer is "we're not sure", that's the thing to find out.
Close
I don't think the AI tools we've come to rely on are about to get expensive. I think they're going to keep changing, and that's the risk worth planning for.
Worry less about the fare. Keep an eye on the route.