Conclusion First: Now is the “Buying Time” for AI.

OpenAI, Google, Meta, Anthropic — giants are competing for supremacy in AI agents with little regard for profit. OpenAI is aiming for market share while projecting a $5 billion (approximately ¥750 billion) loss in 2024, Google is bundling Gemini with its cloud services for almost free, and Meta has even released Llama 3 as open-source software at no cost.

While they are engaged in a bloody battle, what is happening? The cost for those “using AI” is plummeting to the ground.

This isn’t just a story for large corporations. Local SMEs, operating with 10 or 20 employees, have fruits they can harvest right now. Let’s discuss three specific examples.

—

Fruit #1: The Collapse of API Prices — A World Where a “¥3 Million Estimate” Becomes ¥5,000 a Month

First, take a look at the numbers.

When OpenAI’s GPT-4 Turbo debuted in November 2023, it cost $10 per 1M tokens for input. By May 2024, with the release of GPT-4o, that price was halved to $5. Then, in July 2024, the GPT-4o mini dropped to $0.15. That’s a staggering 98.5% cost reduction in just 8 months.

Competitors are following suit. Google’s Gemini 1.5 Flash offers a price of $0.075 per 1M tokens with a 128K context. Anthropic is also gradually lowering its prices. Moreover, Meta’s Llama 3.1 is open-source, meaning if you run it on your own servers, the API cost is zero.

What does this mean in practice?

Two years ago, a manufacturing company that wanted to “automate inquiry responses with AI” received a quote of ¥3 million from a system integrator. This included custom development, server setup, and maintenance costs. Naturally, the proposal didn’t get approved.

Now, if they were to attempt the same thing, what would happen? By using the GPT-4o mini API and a no-code tool like Dify, they could operate for ¥5,000 to ¥10,000 a month. Moreover, the accuracy would surpass that of the top-tier model from two years ago.

¥3 million becomes ¥5,000 a month. This is not just a story of “lower prices”; it represents a structural change where there is now “no reason not to do it.”

Thanks to the zero-profit battles among major vendors, SMEs can benefit simply by “using it.” They should take full advantage of this unprecedented low pricing before these companies end their war of attrition.

—

Fruit #2: Talent Exodus from Large Corporations — “Previously Unattainable Talent” is Now Available

Standard Chartered Bank has announced plans to cut over 7,000 jobs in the next 3 to 5 years due to automation through AI. Meta is also planning to implement layoffs in the thousands from 2024 to 2025, citing performance reasons, while simultaneously reallocating staff to AI departments.

This isn’t just a story in finance and tech. Consulting and IT giants like Accenture, IBM, and SAP are also accelerating organizational restructuring due to AI adoption.

Here, I want to ask SME owners: “I want to hire someone knowledgeable in AI, but they won’t come to a company like ours” — do you think this way?

That common perception is changing.

The talent entering the market due to layoffs from large corporations isn’t just limited to those who performed simple tasks. It includes mid-level professionals who were involved in planning and promoting AI projects, data analysts, and individuals capable of designing business processes. Many of them are tired of the rigid structures of large companies.

Hiring full-time employees is not the only option. Nowadays, side jobs and contract work are the norm. By contracting AI talent from large corporations for project-based work at ¥100,000 to ¥200,000 a month, even local SMEs can afford this.

In fact, one of our clients, a food processing company with 15 employees, contracted a former DX promotion officer from a major manufacturer as a remote advisor for ¥120,000 a month. With their advice, they automated their order and invoicing processes with AI, reducing the staff’s overtime by 40 hours a month.

The timing when large corporations “release talent” is the timing when SMEs can “acquire talent.” This window won’t last long. As the economy recovers, these individuals will return to large companies. If you’re going to act, now is the time.

—

Fruit #3: Democratization of Tools — “Lack of Development Capability” is No Longer an Excuse

The third point may be the most important.

Two years ago, if you wanted to integrate AI into your business, you either needed an engineer who could write Python in-house or had to outsource it. What about now?

• Dify, Coze, Zapier AI — Build AI agents without coding
• ChatGPT Teams, Claude for Work — Monthly plans for team-wide AI usage (around $25-$30 per person)
• Google AI Studio, Amazon Bedrock — Switch between multiple AI models using a GUI
• Llama 3, Mistral, Gemma — Open-source models available for commercial use at no cost

Furthermore, noteworthy is the emergence of the MCP Protocol (Model Context Protocol). This specification, set to be released by Anthropic at the end of 2024, will allow AI agents to connect with external tools (like Google Drive, Slack, databases, etc.) in a standardized manner. Once this becomes widespread, the entire process of “having AI read your business data, make judgments, and execute actions” can be assembled without programming.

Let’s provide a concrete example. A local real estate management company (with 8 employees) built a property inquiry response bot using GPT-4o mini on Dify. The development period was 3 days, with zero outsourcing costs. The CEO created it by following a YouTube tutorial. As a result, monthly phone inquiries decreased by about 30%, and the administrative staff was able to increase property viewings, leading to higher conversion rates.

Once, “development capability” was the exclusive domain of large corporations. Now, with “motivation and three days,” even SMEs can operate AI agents. Knowing this fact can drastically change your competitive edge a year from now.

—

So, What Should We Do?

I have laid out the three fruits. Now, let’s discuss “what to do starting tomorrow.”

Step 1: First, decide on one task to automate.
You don’t have to think big. Choose a small repetitive task like “replying to routine emails that take 30 minutes every day,” “checking invoices at the end of the month,” or “sending initial replies to job applicants.”

Step 2: Try it for free or for less than ¥5,000 a month.
ChatGPT Teams (at $25 a month) is fine. Dify’s free plan is also good. Just start using it. If after a week you think, “This is useful,” continue. If not, try something else. The cost of this trial is now less than one-hundredth of what it used to be.

Step 3: Eliminate one instance of dependency.
The greatest value of automating tasks with AI is not cost reduction. It’s about reducing the number of tasks that can’t function without a specific person. This becomes a lifeline for SMEs. Retirements of veteran employees, sudden absences, and hiring difficulties — all stem from dependency. By transferring knowledge to AI agents, you can create a system that anyone can operate with the same quality. Just that alone can change the resilience of your company.

—

The Attrition War Among Giants is a “Subsidy” for SMEs

Finally, let’s summarize the structure once more.

The AI technologies developed by OpenAI, Google, Meta, and Anthropic, which have invested trillions of yen, are now accessible to SMEs at almost no cost due to zero-profit price competition. Talent that was previously unattainable is now entering the market due to layoffs in large corporations. The evolution of no-code tools allows AI to be integrated into business without needing engineers.

Now, with all three of these factors occurring simultaneously, it is a historic “buying time” for SMEs.

However, this situation will not last. The price competition among large companies will eventually lead to consolidation, and prices will rise. The window for talent in the job market will also close as the economy recovers. Free plans for tools will become paid as the number of users increases.

It’s simply a matter of whether you will pick the fruits while they are still on the ground.

If you have time to think, I encourage you to ask ChatGPT, “List 10 tasks in our company that could be automated.” The answers are already there.