Please Apologize to the Me Who Panicked and Threw Everything at Fable — The Limits of Subscriptions, Seen Through the AI Vendors' Chaos

Last week, I quietly changed the order of my work.

“Let me throw this at Fable while I still can.”

That was my thinking.

Fable was offered as a limited-time trial from the very beginning.

After that, it would fall outside the subscription and move to credit-based billing.

That was explained from the start, so I don’t feel deceived.

So I deliberately routed my heavy work — design reviews, technical investigations — to Fable first.

“Use it while you can.”

I don’t think I was the only developer who thought that way.

But then the situation changed.

The deadline was extended again and again, and it finally settled into a form where on the Max plan you can use up to 50% of your weekly allowance inside the subscription.

As a Max user, I welcome that.

Still, watching this whole scramble, one thing struck me.

Selling AI may be harder than the AI itself.

Anthropic’s Decision Makes Sense

I don’t really feel like blaming Anthropic for this.

The compute cost of the latest models is extremely high.

Subscriptions do have limits on usage time and volume, but even so, continuing to offer the newest model for a flat fee alone cannot be easy.

So the judgment that “Fable is credit-billed” is perfectly understandable for a company.

A business that doesn’t turn a profit cannot survive.

GPUs don’t run on good intentions.

What They Didn’t See Coming Was the Users’ Reaction

But there was probably one thing they didn’t anticipate.

Fable was too useful.

I thought so many times during that period.

Design reviews.

Complex technical investigations.

Improvement proposals based on the entire codebase.

For work like this, there were plenty of moments where I felt, “This really is a notch smarter.”

So even knowing it was a limited-time trial, I thought, “Let me finish the important work before it ends.”

The problem came afterward.

In the official service, it’s credit-billed.

And then the complaints erupted all at once: “I’m paying for Max, and you want even more money?”

Everyone understood, intellectually, that it had been a trial.

Even so, when something whose convenience you’ve already tasted is taken away — and then you’re asked to pay extra on top — resentment surfaces anyway.

So the heart of the anger wasn’t the fact that it was limited-time.

It was this one point:

“It’s a subscription, and I still have to pay more?”

What Even Is a Subscription?

This uproar made me suddenly wonder.

What even is a subscription?

Of course, I understand the provider’s side.

The latest model is expensive.

So they want to charge extra.

As logic, it’s correct.

But users sign up on a feeling that comes before logic:

“I’m paying a lot of money every month.”

Max users especially.

It’s the most expensive plan, so expecting “I get to use the best model” is only natural.

So when you’re told, “Only the latest model costs extra,” you start to feel:

“Then what is this Max plan for?”

And this isn’t only a story about the AI industry.

Video streaming services do it too.

They say, “You’re a subscriber,” and then, “But the newest movies require a rental fee.”

Every time, I think, “Wait — what was a subscription again?”

Of course, as a business it makes sense.

Hit titles carry high licensing fees.

So they cost extra.

The logic holds.

But emotion is a separate thing.

Users feel, “I’m already paying every month — and I still have to pay more?”

I think that was the real essence of the Fable uproar too.

This Episode Should Be a Lesson for Every Vendor

In the end, Anthropic chose a compromise: “Up to 50% inside the subscription on the Max plan.”

It was probably the result of weighing both the users’ dissatisfaction and the reality of compute costs.

But every vendor should have learned one thing from this.

How to sell the latest model.

If you don’t decide those rules clearly from the very start, changing them later is extremely hard.

From now on, maybe this becomes the norm: “The latest model is metered,” and “it gets folded into the subscription a few months later.”

The Problem We Should Solve Before AI

AI has become astonishingly smart over the past few years.

It reasons through complex designs.

It can run long investigations.

It even handles code reviews.

But what this episode revealed is that what the vendors are really agonizing over is not AI’s performance.

How to avoid running at a loss.

How to make users feel the price is fair.

No one has found that answer yet.

AI can now perform world-class reasoning.

And yet, the one question they still don’t seem to have asked the AI is:

“Will users accept this pricing plan?”

Then again, even if they had asked, the answer might have been, “Let’s experiment on the users first.”