What Happens When a 3 Million Yen Job Becomes 50,000 Yen

To get straight to the point: the prices for ‘hands-on work’ are beginning to collapse.

An AI agent designed a RISC-V CPU from a specification of just 219 words. The process took 12 hours. Traditionally, this would have required specialized engineers several months and cost over 3 million yen if outsourced. Now, it costs only tens of thousands of yen in API usage fees and electricity. That’s a cost reduction of over 98%.

Meanwhile, in the accounting world, the AI agent “Pista” has achieved scores surpassing human experts in automated spreadsheet audits, with instances of GPT-4o-class models demonstrating accuracy exceeding that of CPAs (Certified Public Accountants) in accounting tasks.

These two pieces of news may seem to come from different industries, but the underlying structure is exactly the same.

The value of those who can write specifications correctly is skyrocketing, while the value of those who merely execute tasks according to specifications is plummeting.

For small and medium-sized enterprises in rural areas, is this structural change a threat or an opportunity? Let’s consider it from the perspective of outsourcing costs.

—

What Value Remains in a World Where CPU Design Takes 12 Hours?

The topic of discussion this time is the case where an AI agent completed the design of a functioning RISC-V processor with just 219 words—a specification that is less than half an A4 page.

The key point is not that “the AI was excellent.” It is that the structure has been proven: if the specification is correct, the implementation can be completed automatically.

Previously, hardware design followed this flow:

1. Requirement definition (deciding what to create)
2. Architecture design (deciding how to assemble it)
3. RTL design (writing the actual code)
4. Verification and testing (confirming it works)

Among these, steps 3 and 4 consumed vast amounts of labor and time. The monthly cost for specialized engineers ranged from 800,000 to 1,500,000 yen. For a three-month project, labor costs alone would be between 2.4 million and 4.5 million yen. This is the reality behind the “outsourcing costs of over 3 million yen.”

AI agents have nearly entirely replaced steps 3 and 4. What remains are steps 1 and 2, namely, “what to create” and “how to write the specifications.”

The ability to write a specification of 219 words has become the dividing line of value.

—

Why AI Could Outperform Professionals in Spreadsheet Audits

Let’s look at another example. The AI agent “Pista” automatically performs spreadsheet audits—checking formula integrity, detecting anomalies, and identifying reference errors.

Traditionally, this task was performed manually by accountants or accounting staff through visual checks. If there are 100 rows, it might be manageable. But when it comes to 10,000 or 100,000 rows, human concentration cannot avoid overlooking errors. In fact, research data indicates that about 88% of spreadsheets contain some form of error.

What makes Pista interesting is its design to visualize the audit process. The AI not only points out “where the issue is” but also displays the reasoning behind its judgments step by step. Users can review this decision-making process and only perform the final approval.

This is not a case of “dumping everything on AI”; rather, it is a workflow where “AI drafts, and humans review.” As a result, accuracy increases compared to human-only efforts, and the time spent on tasks is significantly reduced.

Here again, the structure is the same. The value of those who can define “what to check” remains, while the value of the “checking tasks themselves” approaches near zero.

—

98% of Outsourcing Costs Disappear—Calculating the Break-Even Point

Now, let’s think about the numbers concretely.

Case 1: Outsourcing System Development

The difference is approximately 3.4 million yen. If there are 10 projects a year, that amounts to a difference of 34 million yen.

Case 2: Monthly Accounting Audits

For small and medium-sized enterprises in rural areas, an annual difference of 1.7 million yen is significant. It corresponds to the bonus of one employee.

Where is the Break-Even Point?

The initial costs associated with utilizing AI agents mainly include the following three:

1. Securing and training personnel who can write specifications: 100,000–300,000 yen/month (training existing employees or utilizing part-time personnel)
2. AI tool usage fees: 10,000–50,000 yen/month
3. Establishing a verification and review system: 200,000–500,000 yen initially

In total, the initial cost is around 500,000–1,000,000 yen, with ongoing monthly costs of about 150,000–350,000 yen. For companies whose traditional outsourcing costs exceed 300,000 yen/month, it becomes profitable from the very first month of implementation.

A monthly cost of 300,000 yen is not uncommon for small and medium-sized enterprises in rural areas when combining website maintenance, system modifications, and accounting outsourcing.

—

“The Ability to Write Specifications” Becomes the Strongest Skill

To summarize the discussion so far:

• What Increases in Value: The ability to define specifications, the ability to articulate “what to create” and “what to check”
• What Decreases in Value: The tasks of executing according to specifications, routine implementation, verification, and auditing

This structure favors small and medium-sized enterprises over large corporations. Why?

In large corporations, the roles of “those who write specifications” and “those who execute tasks” are divided. When the execution department becomes entirely unnecessary, organizational restructuring is required, leading to delays in decision-making.

On the other hand, small and medium-sized enterprises can have the business owner write the specifications themselves. “Our operations are like this,” “We want to automate this”—this on-the-ground sense directly translates into specifications. With no intermediaries, the distance to AI agents is shorter.

Writing a specification of 219 words does not require corporate approvals.

—

So, What Should We Do?

I have three suggestions:

1. List All Current Outsourcing Costs

First, assess the current situation. How much are you paying each month, and for what? Website maintenance, system modifications, accounting services, design production—write it all down.

2. Start Practicing Writing Specifications

The specifications you provide to the AI agent are almost the same as an order form. “What to do,” “under what conditions,” and “when is it considered complete.” If you can articulate these three points, the AI agent will operate. No advanced technical knowledge is required.

3. Start Small

There’s no need to entrust the core system to AI right away. Start with areas where failure won’t be painful, such as spreadsheet audits, automatic meeting minutes generation, or creating simple web pages. The cost is just a few thousand yen per month. Even if you fail, it’s only the cost of one night out.

—

“Hands-On Outsourcing” Quietly Comes to an End

A CPU designed in 219 words and AI outperforming professionals in spreadsheet audits. While this is still a cutting-edge example, it is highly likely that in a year, it will become “normal.”

Technological changes often go unnoticed until it’s too late. When film cameras were replaced by digital ones, many developing labs thought, “We’re still okay.”

What small and medium-sized enterprises in rural areas should do now is not to become experts in AI. It is to become capable of explaining their operations in 219 words.

Companies that can do this will see 98% of their outsourcing costs disappear. And with the freed-up funds, they can take the next step.

While large corporations struggle with organizational restructuring, small and medium-sized enterprises can move ahead. This is a rare opportunity for a turnaround.

Let’s write specifications. Just 219 words.