CODE IN TOKYO

Maybe Anthropic Just Is Not Very Good at Marketing

A Company That May Be Too Honest

Watching Anthropic lately has left me with a strange feeling.

I use Claude Code quite a lot in my daily work, so if anything, I am on the supportive side. It has genuinely helped me, and I still think it is an excellent tool.

Even so, looking at Anthropic’s recent moves, I increasingly find myself wondering:

Maybe this company is not actually very good at marketing.

Anthropic’s defining trait is that it is extremely serious.

It evaluates danger.

It discloses risk.

It verifies safety.

If something looks risky, it stops the release.

As a research attitude, I think this is admirable.

But as a business, there are moments that make me tilt my head.

The company says, “This model may be dangerous.”

Then it says, “So we are stopping the release.”

Then it says, “We have prepared a corrected version.”

At that point, governments and regulators respond by saying, “So it is dangerous?”

From the outside, it is hard not to think:

Wouldn’t they have been better off staying quiet?

Of course, I am not saying companies should simply hide things.

But the structure can make it look as if honesty is being punished.

The Fable Problem

Personally, I think this issue is even more serious.

When I look at reactions from people who tried Fable, the words that stand out are things like:

Then usage limits are applied.

What happens next?

Users go back to the previous model.

The problem is that they have already seen what comes next.

Imagine being taken to a high-end restaurant and then being told, “Starting today, we are eating at the company cafeteria.”

It is not that the cafeteria is bad.

It simply loses by comparison with yesterday.

Humans do not evaluate things in absolute terms. We evaluate them relatively.

If users had never known what was possible, they might have remained satisfied.

But once they have seen the future, they cannot really go back.

Something Feels Missing

This is completely subjective.

I am not talking about benchmarks.

When I use today’s Opus, I sometimes feel that it is a little less intelligent than the previous Opus.

I do not know whether performance has actually dropped.

Maybe it is just my imagination.

Maybe my expectations have risen.

Maybe Fable has become the comparison point.

Still, from a user experience perspective, this matters.

Users are not reading papers.

They use the tool every day.

If they start thinking, “This somehow feels worse than before,” then that becomes the reality of the product.

An Opportunity for OpenAI

Recently, I have started moving some of my work to Codex.

Of course, I still think Claude Code is smarter in many situations.

When design and research are included, it remains strong.

But tool choice is not decided by performance alone.

Is it stable?

Can I use it when I need it?

Can I access it consistently?

Does it avoid disappointing my expectations?

These factors matter a lot too.

Users are not necessarily fans of Anthropic.

I am not either.

I just want to use good tools.

So when disappointment accumulates, the conclusion becomes:

Maybe OpenAI is good enough.

That is how competition works.

Today’s Impression

I still think Anthropic is unquestionably one of the top companies in technical capability.

In fact, I respect it as a group of researchers.

But watching the past few months has reminded me of an obvious truth:

Excellent researchers are not always excellent marketers.

And the most ironic part is this:

Anthropic’s biggest rival may not be OpenAI.

It may be yesterday’s Anthropic.

After seeing Fable, users can no longer return to the past.