Two companies provided quotes, and both were 10 million yen.
The same thing was completed by morning for $200.
This is a story that Yoichi Ochiai recently experienced.
"10 million yen? Can't you do it for 500,000?"
When he asked that, the response was, "500,000 won't work."
So, Mr. Ochiai assigned the project to an AI agent.
When he woke up the next morning, it was done.
It cost approximately $200.
It would take two days to do it yourself. Hundreds of thousands to outsource it.
Have you ever given up on a job because of that?
A project pushed to the edge of your desk. A case you stopped proposing internally after seeing the estimate.
Those price tags are now collapsing.
The reason 10 million yen became $200 is something Mr. Ochiai himself explained.
It is the "reinvention of the wheel."
Most services are just reinventing the wheel with a tiny bit of something new. About 99% of the code is a reinvention, and the cost for that has dropped to a few tens of thousands of dollars, he says.
There is even a story about a C compiler—something incredibly difficult to build—being created with about $20,000 in API charges.
And Mr. Ochiai also said this:
The tools were mostly ready by March. What couldn't be done in December of last year became almost entirely possible by March. This is a huge difference.
99% is a rehash. The cost of that rehashing has dropped. The tools were ready by March.
That is why 10 million yen becomes $200.
It is not magic; it is a repricing.
People taking estimates based on the common sense of a few months ago are looking at outdated price tags.
If you hear about Mr. Ochiai's own daily routine, your perception of speed will change.
In the hour and a half between getting off a plane and arriving at a recording studio, he was able to write 16,000 lines of code while having another conversation.
The amount a human can write in a full day of hard work is about 1,000 lines.
By the time he goes to bed, it reaches 32,000 lines, which is about 32 times faster. Moreover, he feels that by the end of the year, it will be 100 times faster.
What does the screen look like? In Mr. Ochiai's words, he has many agents open on the screen, letting them work simultaneously.
Even at that moment during the recording, two agents were running right in front of him.
If you read this thinking it is just a story for engineers, you will likely lose out.
Proposals. Competitor comparison tables. Reports generated from minutes. Weekly routine materials.
All of these are things that "I could do myself if I had the time, but I outsource because I don't."
The unit price of that time is now collapsing.
In another context, Mr. Ochiai spoke about the cost itself.
The cost to access the same latest AI models is dropping at a rate of about 1/900th per year. So, something like the GPT of a little while ago can now be used for 0 yen.
The duration they can work has also changed.
Around the time of GPT-4, it would fail after about 8 minutes. Now, it can work for about 10 hours in his environment.
If you start it in the morning, it keeps working for 10 hours, and it moves at about 100 times the speed of a human.
Here is my perspective.
Cheap, long-working, and fast.
Try placing these three alongside the story of 10 million yen becoming $200. It starts to look like something that will happen normally from now on, rather than a special success story.
Here is another perspective of mine.
What Mr. Ochiai called 99% was about code.
But what about the proposal you made last month?
You opened a template, swapped the numbers, and adjusted the wording.
What percentage of that was actually new?
Things on the side of "rehashing" will drop in price. Or rather, they have already dropped.
There is another quote from Mr. Ochiai that hits home.
We aren't distributing money through basic income, but right now, we are distributing ability.
Mr. Ochiai views this as a form of "redistribution." Anyone can use it, and if you ask the AI to build a new system, it can. But he also says that most people don't build anything.
And he doesn't stop there.
When the interviewer suggested, "There is a difference in leverage between those who can make it 100x and those who can make it 1000x," Mr. Ochiai replied:
"Exactly. The gap in leverage hasn't been closed. That's why it might widen the disparity even further."
Moreover, extremely high abilities are being distributed at the bottom line. So, he says, this is quite cruel.
Mr. Ochiai described this situation as "merciless."
"You can build a service using Claude Code, so you should just earn money with that too."
The tools are already in your hands. So you should be able to earn, right? That is the era we are in.
The person being told that can't stand it.
However, the situation where that is said has already arrived.
So how do you prepare? The most effective thing here is the paradigm shift Mr. Ochiai discussed.
Until now, humans and machines collaborated to create something.
But now, collaborating with machines to create makes things slower. So, he says, we are moving toward looking at things created automatically and choosing from them.
If infinite songs are produced, choosing which song to pick is harder than creating one.
From the role of the creator to the role of the selector.
And one more thing comes up that sounds familiar.
When writing code with AI agents, Mr. Ochiai says the biggest bottleneck is human approval. Human operation is the biggest bottleneck; it would be faster if humans didn't operate it.
This shouldn't be limited to stories about code.
Approval. Confirmation. Decision-making.
What is stalled is not the work, but usually someone's inbox.
Based on the discussion so far, I will organize the way to request work tomorrow into 5 steps.
The subject will be limited to one thing: the competitor comparison table you make every month.
Step 1. Choose from the side of rehashing.
99% is reinventing the wheel, Mr. Ochiai said. A comparison table where only the numbers change while the format remains the same as last month is the first price tag to be removed from your work.
Step 2. Give it a night.
Ask, "Using last year's format, replace the numbers with this month's and create 3 patterns of comparison tables," and then go to sleep. In Mr. Ochiai's case, the AI works for 10 hours. The time while you sleep is usually enough.
Step 3. Run them in parallel instead of fixing one by one.
If you have one draft perfected and then fix it, you wait every time there is a back-and-forth. Run 3 patterns simultaneously and look at them all together in the morning.
Step 4. Stop mid-way confirmations.
Don't let it ask "Is this okay?" for every item. Let it run to the end before you look. As long as you insert a confirmation every time, you are the cause of the traffic jam.
Step 5. In the morning, choose one from the three options.
Don't recreate it; choose it. Move from the creator's chair to the selector's chair.
If it works for the comparison table, try applying the same five steps to reports generated from minutes.
I don't guarantee that 10 million yen will become $200. But there should be at least one project at hand where the digits will change.
There is also a reason to hurry.
Mr. Ochiai says that engineering coding tasks will completely disappear within this year.
The ability to write programs will almost vanish, and anyone will be able to create. He says it is inevitable that entry-level engineering jobs will disappear.
This isn't a story for next year. It is being discussed as something happening within this year.
Finally, because I don't want to end by just instigating fear, I will share the best story Mr. Ochiai left behind.
Development will become easier, so anyone will be able to make a Pokemon game.
But nobody wants to play a knock-off Pokemon.
Not everything will lose its value.
Things that can be duplicated will have their price tags broken. Rehased proposals, template materials, and reinventing the wheel will all fall to the $200 side.
But things that cannot be duplicated will maintain their price.
The IP you have nurtured. The judgment only you can make. The trust you have built up. The reason someone wants to ask you specifically.
Mr. Ochiai also says that even if the AI bubble bursts, the world after it will not.
The people who laid the infrastructure for railroads, canals, and fiber optics didn't make money. But thanks to them becoming cheap, the Industrial Revolution was completed, and YouTube emerged.
After the oversupply, the real show begins.
So the fact that the price of tools is dropping is probably not a scary story.
What is scary is this:
AI agents are being distributed to people all over the world, and in terms of labor productivity, everyone is very close to having overwhelming productivity.
Yet, the reason new things aren't coming out might be because there are extremely few people trying to create new things with them.
Mr. Ochiai says this and calls it "the essential problem that humans are currently being confronted with."
You are also on the side that received the distribution.
So tomorrow, you only need to decide one thing.
How much of your work will you throw into the $200 side, and what part will you keep in your own hands?
Only those who clearly know the parts that won't become a knock-off can protect their price tag.
