The Trap of "Blazing-Fast Development": Expectation Control in the AI Era

When Speed Becomes Normal

Recently, when I develop with AI, I sometimes feel a little scared.

Work that used to take half a day can now be done in ten minutes.

Research that I expected to take a full day can be wrapped up before lunch.

For a programmer, it feels like a dream.

But there is also a slight danger hidden in that speed.

Suppose a client asks for a feature, and I quickly implement it with the help of AI.

At first, the client is happy.

“Wait, it is already done?”

I also feel a little proud.

But humans get used to things.

And expectations always rise.

• Stage 1: “Amazing!” (surprise)
• Stage 2: “Well, this much is normal, right?” (familiarity)
• Stage 3: “Isn’t this slower than last time?” (dissatisfaction)

Someone who was treated like a miracle worker yesterday becomes ordinary today, and may be treated as incompetent tomorrow.

It is a rather unreasonable world.

What We Need Is Expectation Management

That is why, recently, even when a task can be done in 30 minutes with AI, I tend to say something like:

“I will get back to you this afternoon, including confirmation.”

This is not because I am slacking off.

There is testing.

There is specification checking.

There is documentation.

And above all, the world contains a terrifying thing called the unexpected.

Just because AI has become faster does not mean uncertainty has disappeared from projects.

If anything, implementation has become abnormally fast, so the gap between speed and expectations has become more visible.

Implementation and Completion Are Different

What AI is good at is writing code.

The confirmation, review, and quality assurance that come after that are still human work.

So in reality:

“The code is done”

and

“The work is done”

are two different things.

However, when you share your screen and show AI generating code, people often say:

“So it is basically finished, right?”

No, that is where the real work begins.

Becoming Neither a God nor an Incompetent Person

In the end, I think what matters in the AI era is not only how to use AI, but how to manage expectations.

There is no need to force yourself to become a god.

If you become a god, people will expect divine work from you next time.

At the same time, being seen as incompetent is obviously not ideal either.

Leave a reasonable amount of margin, and keep delivering stable results.

That is probably the best way to survive for a long time.

I write all this as if I know what I am talking about, but this article itself was also written with AI’s help in about ten minutes.

If readers start saying next time:

“You can write blog posts in ten minutes too, right?”

then the argument of this article will have been perfectly proven.