Automation in 325 Lines of Python: A Structural Shift Turning a 300,000 Yen Outsourcing Job into a 500 Yen Monthly Expense

Work That Was Outsourced Now Completed in 325 Lines

How much does your company pay each month for “work that someone else does”?

Creating summaries of industry news, transcribing data into PDF forms, organizing and assigning daily tasks—these are all tasks that fall into a gray area of being “too cumbersome to do myself, but not worth outsourcing.” As a result, office staff either handle them manually, outsource them for tens of thousands of yen a month, or simply choose not to do them at all.

This structure is now fundamentally changing.

Recently, while looking at open-source projects, I noticed a common theme: “Individuals are creating hundreds of lines of code over the weekend.” A tool that automatically summarizes AI news in 325 lines of Python. “SimplePDF Copilot,” which automates PDF form entry in the browser. An app that reads calendars and ToDos to reorganize schedules. All of these have functionalities that would have cost between 1 million and 3 million yen if you had sought estimates from development companies just a few years ago.

So, what is the cost of running these tools now? This is the crux of the matter.

—

Breaking Down the Costs—The Breakdown of “500 Yen a Month”

Let’s break down the costs of the 325-line Python news summarization tool.

The time required for development is about 1 to 2 days for someone with basic Python skills. If you lack those skills, you could consult Claude or ChatGPT while progressing, which might take around 3 to 5 days.

What if you outsourced this? If you asked someone to summarize daily news and compile it into a report, it would cost you 30,000 to 50,000 yen a month. If you ordered it as a tool from a system development company, the initial cost would be between 500,000 and 3 million yen, plus monthly maintenance fees.

500 yen a month vs. 30,000 yen a month. A 60-fold difference.

The SimplePDF Copilot for automatic PDF form filling is even more extreme. Since it operates client-side (in the browser), server costs are nearly zero. Its design ensures that data is not sent externally, making it easy to use for processing forms that contain personal information. Application documents for government submissions, reports for clients, various internal notifications—tasks that were previously done manually are now “automatically completed.”

What’s important here is not just the fact that costs have decreased. The decrease in costs has changed the criteria for deciding whether to do something or not.

—

The Reasons for Not Doing It Are Disappearing

If it costs 300,000 yen a month, then it’s not worth doing. That’s obvious. Considering the profit margins of small and medium-sized enterprises, they cannot afford to spend 300,000 yen on a task that is merely “convenient to have.”

But what if it costs 500 yen a month? There’s no reason not to try.

The disappearance of this “hurdle to try” represents the biggest structural change for small and medium-sized enterprises.

Large companies have historically automated operations through system investments of tens of millions of yen. They implement ERP systems, adopt RPA, and have dedicated IT departments manage them. Small and medium-sized enterprises lacked the capacity for that. As a result, manual work persisted, and reliance on specific individuals became the norm, leading to a situation where “things don’t run without that person.”

However, this structure is now being flipped.

Large companies, because they have implemented massive systems, cannot be agile. When a new LLM is released, it can take six months to verify compatibility with existing systems. Security reviews take three months. Approval processes take a month. As a result, we find ourselves in a laughable situation where “even though GPT-4o has been released, we are still verifying GPT-3.5.”

On the other hand, small and medium-sized enterprises are different. If the president thinks, “This looks useful,” they can try it the next day. There’s no need for approval to run a 325-line script. If it doesn’t work, they can discard it. It’s only 500 yen a month.

The speed of decision-making multiplied by the low cost of implementation. This becomes the weapon for small and medium-sized enterprises.

—

“Dependency” Is Both the Greatest Risk and the Greatest Opportunity

However, there is one pitfall here.

The issue of “not having anyone in-house who can write Python.”

For someone who can code, 325 lines of code is easy. But for someone who cannot code, 325 lines are just as impossible as 3,250 lines. Ultimately, they may still end up relying on someone outside, and the cost structure remains unchanged—you might think.

However, change is also happening here.

From the latter half of 2024, the accuracy of AI coding assistants has dramatically improved. With Claude, ChatGPT, or GitHub Copilot, it’s no longer necessary to “know how to write Python.” The requirement has shifted to being able to “explain what you want to do in Japanese.”

In fact, there’s a case from my team where we tried this. We had a staff member with no programming experience use Claude to create a script that “fetches new information from a specific website every morning, summarizes it, and posts it to Slack.” It took four hours, and they produced something that works. It’s not perfect, but it functions. And from the next day, summaries started arriving every morning.

This doesn’t mean that the “cost of programming” has decreased. It means that the “cost of systematization” has decreased.

The essence of dependency is that “the know-how exists only in that person’s head.” The task of translating that into a reproducible form, like code, could only be done by experts until now. But now, the people who know the business best can use AI to systematize it themselves.

This is significant. There’s no need to play a game of telephone with an outsourcing partner or IT department to convey, “I want to do this.” The people who know the business can directly create the automation systems. The loss from the game of telephone becomes zero.

—

“So, What Should We Do?”

Let’s move away from abstract discussions. Here are three things you can do starting tomorrow.

1. Choose One “Manual Task You Repeat Every Month”

Don’t try to automate everything at once. Start with one. A task you do every month that follows a set pattern and is cumbersome. Transcribing invoices, aggregating reports, standard email replies—anything is fine.

2. Consult ChatGPT or Claude About “Automating This”

The prompt can be simple. “Every month, I transcribe data from this column in Excel into this PDF form. Please write a Python script to automate this.”—with just this, you’ll receive working code back. It doesn’t have to be perfect. It just needs to work.

3. Verify Whether It Can Operate Within 1,000 Yen a Month

Once it’s working, the next step is to check the costs. If you run it in a serverless environment (like AWS Lambda or Google Cloud Functions), it will usually stay within a few hundred yen a month for the workload of a small or medium-sized enterprise. If it exceeds 1,000 yen a month, reconsider the design or whether that task really needs to be automated at all.

—

The Real Question Is Not “Technology”

Finally, just one thing.

Whether you can write 325 lines of Python is no longer the issue. AI can write it for you. Server costs are also 500 yen a month. The technical barrier has effectively disappeared.

The only remaining hurdle is the “ability to identify where in our operations there are repetitive tasks that can be automated.”

This is something AI cannot find. Only the people who do that work every day can understand it.

There’s no need to wait for technological evolution. Everything is already more than sufficient. What’s lacking is simply the habit of thinking, “Could this be automated?”

What can you automate for 500 yen a month? I hope you start asking yourself this question from today.