Code Writing Speed Has Increased 100 Times. So, What Has Changed?

An engineer has stopped using AI coding. The reason is simple: “Even if I write code 100 times faster, productivity was zero.”

This may sound like an extreme statement. However, when we calmly observe what is happening on the ground, it is no laughing matter. The speed at which AI generates code has indeed skyrocketed. With GitHub Copilot, dozens of lines of code can be produced in seconds. If you throw specifications at ChatGPT, it returns seemingly appropriate functions. The sensation of ‘100 times faster’ is not an exaggeration.

But that’s not the question. “Was the code that was written quickly usable as it was?” — That’s the crux of the matter.

When the Cost of ‘Writing’ Decreases, the Cost of ‘Fixing’ Becomes Apparent

The reality described by that engineer is as follows.

The code generated by AI appears to work at first glance. It passes tests. However, as soon as it enters the production environment, unexpected bugs emerge. Edge case handling is inadequate. There is a lack of consistency with existing code. Security holes lurk within.

As a result, code written by AI in five minutes requires a human to spend three hours reviewing, correcting, and retesting it. This cycle repeats every day.

Let’s do the math. Traditionally, if it took a human five hours to write a certain piece of code, using AI shortens the ‘writing’ process to five minutes. However, reviewing and correcting takes three hours. Subtracting, that totals three hours and five minutes. It has indeed become shorter. But it is far from the dramatic efficiency imagined from ‘100 times faster.’ Moreover, reviewing and correcting require skills that are more advanced than writing the original code.

As the cost of ‘writing’ code approaches nearly zero, the costs of ‘reading,’ ‘judging,’ and ‘fixing’ code have risen relatively. This is the mechanism that leads to zero productivity even at 100 times faster.

What Happens When Costs Decrease

This structure is not limited to coding.

The cost of writing articles has decreased. Throw it at AI, and a 3,000-word piece can be generated in 30 seconds. However, it is humans who must judge whether “this article is accurate,” “is it valuable to readers?” and “are the facts correct?” The time required for that judgment remains unchanged. In fact, with AI generating large volumes of text, the burden of judgment has increased.

Design is the same. Using image generation AI, you can produce 100 or 200 options for banners or logos. However, it is humans who decide “which one fits the brand” and “does this color resonate with the target audience?” The more options there are, the greater the burden of selection.

In other words:

AI has dramatically reduced the cost of ‘tasks.’ However, the cost of ‘judgment’ has not decreased. In fact, it has increased.

And in the market, while the prices of things with reduced costs are falling, the prices of things whose costs do not decrease are rising. This is a fundamental principle of economics.

The Price of ‘Work That Should Be Done by Humans’ Is Rising

Let’s look at the numbers.

The unit price for simple coding projects on crowdsourcing platforms has clearly dropped over the past two years. “Coding a single landing page” used to cost between 50,000 to 100,000 yen. Now, because AI can create a rough outline, some people are willing to do it for 20,000 to 30,000 yen.

On the other hand, the unit prices for jobs like “design review of existing systems,” “organizing technical debt,” and “clarifying ambiguous requirements in upstream processes” have risen. Reports suggest that the hourly rates for freelance senior engineers have increased by 1.2 to 1.5 times compared to five years ago. Positions that used to pay 800,000 yen per month are now paying between 1 million to 1.2 million yen.

Why? Because AI has become capable of substituting ‘writing,’ the scarcity of people who can decide ‘what to write’ and ‘judge whether what has been written is correct’ has increased.

This is not just a coding issue.

• AI can now create sales materials → The value of salespeople who can judge “what to propose to this customer” has increased.
• AI has automated accounting entries → The value of accountants who can notice “anomalies in these numbers” has increased.
• AI can now write recruitment texts → The value of interviewers who can determine “whether to hire this person” has increased.

The same structure is occurring across all professions.

This Is an Opportunity for Small and Medium Enterprises

Now we get to the main point.

Large corporations are investing millions of yen annually in AI tools to advance large-scale automation. However, small and medium enterprises do not need to imitate this. In fact, they should not.

The strengths of small and medium enterprises are their “speed of judgment” and “closeness to the field.” The president is observing the field. They can see the faces of customers. Therefore, they can make judgments about “whether this decision is correct” faster and more accurately than large corporations.

In an era where AI makes ‘tasks’ cheaper, what small and medium enterprises should do is clear.

1. Delegate tasks to AI, but do not ‘throw it all away.’

The monthly fee for AI coding tools is around 2,000 to 6,000 yen. GitHub Copilot costs $19 per month (about 3,000 yen), and ChatGPT Plus costs $20 per month (about 3,000 yen). That totals 40,000 to 70,000 yen annually. If this can cut the time for the ‘writing’ process in half, it pays off.

However, it is dangerous to use what comes out ‘as it is.’ Humans must always review it. Cultivating ‘people who can review’ within the company is the essence of utilizing AI.

2. Invest in training ‘judgment-capable personnel.’

Spending 300,000 to 1 million yen annually on “AI tool usage courses” is a waste. The way to use tools changes in six months. Instead, invest in skills like “the ability to organize requirements,” “the eye to assess quality,” and “the ability to articulate customer challenges” — these judgment skills.

Specifically, request monthly reviews from external senior personnel (5,000 to 10,000 yen per month), establish a habit of verbalizing “why this judgment was made” in internal study sessions, and require young employees to accompany customer hearings. Such down-to-earth measures will yield returns more reliably than introducing AI tools.

3. Systematize ‘the record of judgments.’

The biggest cause of dependency on individuals is that “the reasons for those judgments” are not recorded. Document the judgment criteria of veterans. This can be supported by AI right now. Have veterans speak, transcribe it with AI, and organize it. This can be done with tools costing a few thousand yen per month.

Once the judgment criteria are verbalized, even young employees can replicate them. This is the true utilization of AI in small and medium enterprises.

So, What Should We Do?

Let’s summarize.

• AI has made ‘tasks’ 100 times faster. However, ‘judgment’ has not become faster.
• The price of tasks has decreased, while the price of judgment has increased.
• What small and medium enterprises should invest in is not AI tools, but ‘judgment-capable people.’
• AI tools should be used for a few thousand yen per month. There is no need for multi-million yen implementation projects.
• Verbalizing and systematizing judgment criteria will prevent dependency on individuals and enhance organizational competitiveness.

What the engineer who stopped using AI coding realized is the obvious fact that “being able to write quickly” and “being able to create the correct thing” are two different matters.

This obvious truth is now being reaffirmed across all professions. Whether it’s code, text, design, or sales materials, in an era where the cost of ‘creating’ approaches zero, the price of ‘people who can decide what to create’ is rising.

What business owners of small and medium enterprises should do is not to implement expensive AI tools, but to understand “who is making judgments in our company.” Turning that person’s judgment criteria into an organizational asset is where it starts.