The World Where One Engineer’s Monthly Salary is 3,000 Yen

Let’s get straight to the point.

By 2025, using AI coding tools, development tasks that previously cost between 500,000 and 1,000,000 yen for outsourcing are becoming a state of being ‘self-completing’ for just a few thousand to tens of thousands of yen a month.

This is not just a story for large corporations. For small to medium-sized enterprises in rural areas—companies with 10 employees and zero IT engineers that want to ‘streamline operations’ or ‘create simple internal tools’—this represents the most significant change.

We no longer need to ask, ‘What can AI do?’ The question now is:

‘How much per month will it do for us automatically?’

We will compare the capabilities of three tools: OpenAI’s Codex, Moonshot AI’s Kimi K2 (Kimi Code CLI), and Anthropic’s Claude Code, focusing on their cost and practicality.

—

What’s Different Among the Three Tools

OpenAI Codex—The Experience of ‘Self-Completing Asynchronously’

Codex is a cloud-based coding agent from OpenAI released in May 2025. It can be used directly from ChatGPT, eliminating the need to open a terminal.

Its most significant feature is asynchronous execution. You can issue commands like ‘Fix this bug’ or ‘Write a script to read this CSV’ and it will complete the tasks while you work on something else. It connects to GitHub repositories and even creates pull requests automatically.

• Monthly Cost: Included in ChatGPT Pro ($200/month, approximately 30,000 yen). It is also available with limitations on the Plus plan ($20/month, approximately 3,000 yen).
• Strengths: Bug fixing, test generation, refactoring, automatic pull request creation.
• Weaknesses: Struggles with tasks dependent on local environments or external APIs due to execution in a sandbox environment.

Notably, it can be used for $20/month (approximately 3,000 yen) with the Plus plan. Tasks that previously cost 50,000 to 100,000 yen for minor fixes could now potentially be completed for a flat fee of 3,000 yen per month.

Kimi K2 (Kimi Code CLI)—The Shock of ‘Free Open Source’

Released in July 2025 by Moonshot AI, Kimi K2 is a 1 trillion parameter MoE (Mixture of Experts) model that has achieved benchmark scores comparable to GPT-4o and Claude 3.5 Sonnet in coding performance.

Moreover, Kimi Code CLI is available as open source for free, under the Apache 2.0 license.

• Monthly Cost: The tool itself is free. For API usage, it charges $0.6 per million input tokens and $2.0 for output, making it one of the lowest in the industry.
• Strengths: Code comprehension, bug fixing, feature additions. Operates interactively via the terminal.
• Weaknesses: Requires terminal operation as there is no cloud UI, posing a barrier for non-engineers.

To put costs into perspective, even with around 50 code generation tasks per day, it would only cost a few hundred to a few thousand yen per month. The price disruption caused by this AI from China is reminiscent of DeepSeek.

However, to be frank, it cannot be used by those who are not comfortable with the terminal. This is the biggest bottleneck for small to medium-sized enterprises.

Claude Code—The Completeness as a ‘Professional Partner’

Anthropic’s Claude Code, released in February 2025, is a terminal-based AI coding agent. It utilizes Claude 4 (Opus/Sonnet) as its backend and makes modifications and suggestions based on an understanding of the entire codebase.

• Monthly Cost: Available through Claude Pro ($20/month, approximately 3,000 yen). The Max plan ($100/month, approximately 15,000 yen; $200/month, approximately 30,000 yen) allows for more token usage.
• Strengths: Understanding large codebases, complex refactoring, multi-file editing.
• Weaknesses: High token consumption, and the Pro plan can easily reach daily usage limits.

Claude Code’s strength lies in its depth of context understanding. It can respond to high-level abstract instructions like, ‘Read the entire project and identify security issues.’ Its score on SWE-bench is 72.7% (as of July 2025), placing it among the top performers in autonomous coding benchmarks.

—

Cost Comparison: What Can You Do for How Much Per Month?

—

So, Which Should Small to Medium-Sized Enterprises Choose?

Let’s organize this into three patterns.

Pattern 1: Companies Without Engineers

→ Codex (ChatGPT Plus/Pro) is the only choice.

The reason is simple. It can be used from the ChatGPT interface, eliminating the need to open a terminal. By issuing commands in Japanese like ‘Write a script for this task,’ code is generated, and pull requests are created on GitHub.

You can start for as little as 3,000 yen per month. For companies that previously outsourced ‘minor tools’ for 300,000 yen, the cost is reduced to 1/100.

Pattern 2: Companies with 1-2 Engineers

→ Claude Code (Max) is the best choice. Use Kimi K2 to reduce costs.

If you have engineers, terminal operations are not an issue. Claude Code’s ability to ‘understand the entire codebase’ shines when it comes to modifying existing systems or adding features.

One engineer using Claude Code can feel like 2-3 times the productivity. If 15,000 yen per month can handle work worth 500,000 to 1,000,000 yen in labor costs, the ROI is clear.

Routine small tasks can be routed to Kimi K2’s API (a few thousand yen per month) to optimize costs.

Pattern 3: Companies That Want to Minimize Costs

→ Kimi K2 (API pay-as-you-go).

Charging $0.6 per million input tokens is a bargain. Considering that similarly performing GPT-4o costs $2.5 to $5.0, the cost is 1/4 to 1/8. However, personnel capable of terminal operations are necessary.

—

What You Really Need to Consider

There’s something more important than choosing tools.

‘When development costs have dropped to 1/10 to 1/100, what will you create?’

Things that small to medium-sized enterprises have given up on due to budget constraints—automation of customer management, automatic generation of estimates, optimization of inventory, internal knowledge searches. Such tools will become a state of being ‘self-completing’ for just a few thousand yen per month.

Large corporations are implementing massive SaaS solutions, paying tens of millions of yen annually. Small to medium-sized enterprises can create custom tools for just 3,000 yen per month using AI coding tools. This is not a matter of ‘mimicking large corporations.’ It’s precisely because they are small and agile that they can do this.

Three years ago, outsourcing website creation cost 500,000 yen. Now, if you instruct AI, it can take shape in a day. The same transformation is about to happen across entire business systems.

Will you think, ‘It’s still too early,’ or will you ‘give it a try first’?

For just 3,000 yen a month with ChatGPT Plus, I encourage you to automate one tedious internal task. Once you experience the ‘self-completing’ state, you won’t be able to go back.