CODE IN TOKYO

No Matter How Much You Improve Your Setup, Paying For Fable 5 Is Still Stronger

Trying Fable 5 For A Limited Time

Right now, I am using Fable 5 for a limited period.

I do not use it regularly. It has been temporarily opened to Claude Code users, so I am taking the opportunity to try it in real work.

As I wrote recently, it is clearly smart. It understands intent without needing many detailed instructions, and I do not have to list as many guidelines such as “write this too” or “be careful about that.”

But while using it, I started thinking about something else.

What was all the effort I put into improving my Claude Code environment?

I organized CLAUDE.md, wrote ADRs, created Skills, and reviewed the project structure so the AI would not get lost. I also cleaned up old documentation.

Of course, all of that still has value.

But when the model becomes one level smarter, it can simply jump over many of the gaps that humans were working hard to fill.

That feels a little sad.

The AI World Feels Like A Mobile Game

Using Fable 5 reminded me of heavy spenders in mobile games.

Free-to-play users spend hours thinking through strategies, making the best use of the characters they have, and trying to defeat a strong enemy.

Next to them, a paying user pulls a powerful new character from the latest gacha and simply beats the enemy by force.

AI is gradually starting to feel similar.

A new model can jump over the environment you spent months building, as if it was nothing.

“What was the environment I spent half a year creating?”

That is the feeling.

But if you are using AI for work, there is no point in being stubborn about it.

For Work, Paying Is Often Faster

If it is a hobby, it can be fun to use free models creatively.

Researching prompts and saying, “This prompt brings out better performance,” is perfectly fine if that research itself is the goal.

But at work, using AI is not the goal.

The goal is to produce the final deliverable.

If paying a few extra tens of thousands of yen per month saves a large amount of work time, it is often faster to pay.

At least for freelancers, your time has a price. If you spend hours tuning prompts just to save on AI fees, the priorities are backwards.

Until recently, the term “prompt engineering” was very popular.

How you give instructions still matters. But I feel it is becoming harder to cover the performance gap between models through human ingenuity alone.

It is like trying to compete with a sports car on a highway by improving your driving technique in a small economy car.

Skill matters, but if the performance gap is too large, you will simply lose.

So What Should Humans Do?

Recently, I spend more time thinking about how to divide work than about how to instruct AI.

Simple tasks are enough for Codex. Design and research can go to Claude’s standard models. For truly difficult work, I use a higher-end model such as Fable 5.

It would be easy to send everything to the best model. But after using Fable 5 for a day, I noticed another problem.

Because it is smart and proactive, it consumes the allocated tokens very quickly.

The more capable the employee, the higher the salary.

That is obvious, in a way.

In the end, the human job is increasingly becoming the work of judging task difficulty and assigning each task to the right AI.

It is a little like the old role of a project manager.

The difference is that these workers operate 24 hours a day, do not complain, and become visibly smarter when you pay more.

Is The AI Era Still About Money?

AI allows individuals to compete with large companies.

I hear this often, and I basically agree.

At the same time, the difference in capital is already entering the AI world.

Someone who spends a few thousand yen per month and someone who spends tens of thousands of yen per month have access to different token limits and different models. Companies can put even larger amounts of money into APIs.

It may have been too early to say that AI has made everyone equal.

In the end, people who can pay for stronger models have an advantage.

Using Fable 5 for a limited time made me feel that reality very clearly.

It is indeed smart.

A higher-level model can overpower the work of environment setup and prompt refinement.

And when this limited period ends, I will probably return to my usual models.

The reason is simple.

Even in the AI era, the thing that eventually pulls humans back to reality is the bill.

Contact

I provide system development and technical consulting using AI, AWS, and Claude Code.

Contact form: https://holly-money-e94.notion.site/390ff30cf18c8086a676fe630d171873