Conclusion

Let’s get straight to the point. The battle between “300,000 yen outsourcing vs. 20,000 yen a month” has already been decided.

For small and medium-sized enterprises, investing 3 million yen to develop a business system is akin to stepping off a cliff. Moreover, it takes 3 to 6 months to deliver, and additional costs pile up with every specification change.

Now, with OpenAI’s Codex (approximately 20,000 yen per month), even employees without programming experience can set up a prototype of a business system in just a few days. The annual cost is around 240,000 yen, less than one-twelfth of outsourcing.

The fact that “costs have changed by more than one digit” is impactful enough. However, what we really need to consider is what happens next. When costs drop dramatically, what changes, and what occurs?

What Can Codex Do?

OpenAI’s Codex has significantly evolved into a coding agent that can be used directly from ChatGPT as of the May 2025 update. Here’s what it can do:

• It writes code, tests it, and even creates pull requests based on natural language instructions.
• It directly integrates with GitHub repositories, allowing it to understand and modify existing codebases.
• It can handle multiple tasks in parallel, fixing bugs or adding features while humans work on other jobs.
• It supports major programming languages like Python, JavaScript, and TypeScript.

This is a different dimension from previous “code completion tools.” If you articulate what you want to achieve, Codex will handle the design, implementation, and testing in one go.

For example, if you instruct it to “create a web app that imports customer lists in CSV format and automatically generates monthly sales reports,” a prototype can be operational in a matter of minutes to hours. Previously, just defining requirements with an outsourcing partner would take two weeks.

What Will Project Solara Change?

Microsoft’s Project Solara is also noteworthy. This is an Android-based “Agent OS” that allows AI agents to operate natively on devices.

The key point is that AI has been elevated from “features within apps” to “the OS itself.” New form factors, including desktop and badge-type devices, have been announced, along with security features like facial recognition.

For small and medium-sized enterprises, the implication is simple. The era of needing special setups to use AI is over. Once you turn on the power, AI will be running. This will become the norm.

Codex is destroying the “cost of development,” while Project Solara is dismantling the “barriers to implementation.” These two developments are happening simultaneously.

Calculating the Breakeven Point with Concrete Numbers

Let’s consider the actual figures.

Case 1: Developing a Business System Through Outsourcing

• Initial development cost: 3 million yen (market rate for a medium-sized business system)
• Maintenance and operation cost: 50,000 yen per month (600,000 yen annually)
• Specification changes and additional development: 500,000 to 1 million yen annually
• Total for the first year: approximately 4.1 to 4.6 million yen
• From the second year onward: 1.1 to 1.6 million yen annually

Case 2: In-House Development Using Codex and Internal Staff

• Codex usage fee: approximately 20,000 yen per month (240,000 yen annually)
• Learning cost for the responsible person: the first 1 to 2 months (about 20% of working hours)
• Cloud infrastructure cost: 5,000 to 10,000 yen per month (60,000 to 120,000 yen annually)
• Total for the first year: approximately 300,000 to 360,000 yen + part of personnel costs
• From the second year onward: 300,000 to 360,000 yen annually

In a simple cost comparison, there’s a difference of about 4 million yen in the first year. As years go by, this gap will only widen.

Moreover, what’s crucial is that with outsourcing, “additional costs and waiting times arise with every specification change.” In in-house development, if you think on Friday, “I want to change this,” you can instruct Codex, and by Monday, it’s operational. The speed difference is greater than the cost difference.

Is the “Quality Risk” Really a Risk?

There are certainly concerns that “systems created by amateurs may have low quality.” However, I want to question this.

Was the quality of the systems developed through outsourcing really high?

In the field of small and medium-sized enterprises, you often hear stories like these:

• The delivered system didn’t fit the operational needs, so they ended up reverting to Excel.
• When they requested a specification change, they were quoted an additional 800,000 yen and had to endure using it as is.
• The representative from the development company left, and maintenance became impossible.

The “quality” of outsourcing is not about the beauty of the code. It’s about whether it can be continuously used in the field.

When developing in-house with Codex, the initial prototype may be rough around the edges. However, you can reflect the voices from the field the next day. You can fix it while using it. “Delivering a perfect system all at once” is less effective than “improving a 70-point system every week” in fitting the needs of the field.

This is difficult for large companies to replicate. In organizations where decision-making requires five levels of approval, they can’t say, “We’ll fix it next week.” The agility of small and medium-sized enterprises becomes a competitive advantage.

The Real Risk is the “Risk of Not Using It”

More serious than quality risk is the “risk of falling behind this change.”

If competitors automate their operations with Codex and reduce the time to create estimates from 30 minutes to 3 minutes, even with the same labor costs, the number of cases they can handle will increase tenfold. They can win in price competition and response speed.

In this situation, if you say, “We’re still observing,” you might find that the competition has already been decided. It’s as if it has “ended on its own.”

So, What Should We Do Starting Tomorrow?

Let’s move beyond abstract discussions and outline three concrete actions.

1. First, try automating one “tedious task” in-house with Codex.

There’s no need to create a core system right away. Start with mundane tasks like daily report aggregation, automatic invoice generation, or formatting inventory data. If you create a success story, you can move on to the next step.

2. Have the person who knows the business best, not just the “IT expert,” handle it.

The essence of Codex is that it can “directly convert business knowledge into a system.” A person who intuitively understands “where the waste is in this business” can create a far superior system than someone with programming knowledge.

3. Secure the 20,000 yen as a “budget for experimentation.”

Even if you fail, it’s just 20,000 yen a month. 240,000 yen annually. This is an amount that small and medium-sized enterprises can afford. With this amount, you can verify whether AI development suits your company. If it doesn’t fit, you can stop. Unlike the 3 million yen for outsourcing, the withdrawal cost is nearly zero.

What Happens After Costs Drop

Finally, let’s discuss something a bit broader.

As tools like Codex become widespread, the very value of the act of “system development” will change. Once, printing was something you relied on specialized vendors for, but with the proliferation of printers, anyone can do it. The same is about to happen in business system development.

What will increase in value at that time? The power to decide “what to create.” Not the technology, but the problem setting. Not the code, but the understanding of the business.

Business owners and employees in small and medium-sized enterprises know their company’s challenges better than anyone else. A new era is coming where this knowledge becomes the greatest weapon.

3 million has become 20,000 a month. The question is whether to use that 20,000.