Hi everyone. Today, I would like to share some thoughts about an OpenAI Japan-related event I attended in Tokyo over the past two days. After listening to their presentations, I had mixed feelings. To be honest, I feel I need to put them into words. Let me start with the conclusion. Frankly, I felt somewhat concerned. As someone who is always open to new career opportunities, I had long admired OpenAI Japan. However, after this event, much of that excitement has faded. So, let me explain why.

The Only Clear Positive: Everyone Was Extremely Kind

There were three sessions at the event, and I attended the first and the last ones, both related to OpenAI Japan. Let me begin with the positive side. The people from OpenAI Japan were genuinely kind and polite. In the tech industry, we sometimes meet people who act as if they are superior to everyone else. But I did not feel that at all from the OpenAI Japan team. They communicated with great emotional intelligence. Their tone was calm, respectful, and pleasant. From the leadership team to the managers and the FDEs, everyone was courteous and professional. However... To be honest, I struggled to find many other clear positives. If I had to mention one more thing, I would say the language barrier may not be as intimidating as I had imagined. Their global roles require both English and Japanese, but one of the foreign leaders spoke Japanese extremely fluently, even better than I do. That gave me the impression that language may not be the biggest obstacle after all. Now that I have covered the positive side, I would like to share my more critical observations.

Part One: New Technology, But an Old Story?

The first session focused on OpenAI FDE, or Forward Deployed Engineer. They presented it as a revolutionary technical upgrade and showed a collaboration case with a major Japanese enterprise. The demo looked impressive. One example they highlighted was converting an old COBOL system at a large Japanese company into Java using Codex. The demo showed COBOL code being captured and quickly transformed into Java. The visuals were flashy, and the numbers moved rapidly on the screen. They described this as a revolution from legacy code to modern code.

But this is exactly where I felt uncomfortable. In my view, the real value of an FDE, and the key to the success of the next-generation FDE model, should be about creating something from zero to one. By “zero to one,” I do not mean rewriting old software in a new programming language. That does not change the company’s business logic. It does not transform the business itself. It is essentially a code translator.

When I entered the industry around 2005, Java was promoted in a very similar way. We were told that, unlike C++, we no longer needed to worry about pointers or memory management, and that we could write logic in a more human-like, object-oriented way. At that time, COBOL-to-Java conversion tools were already everywhere. Now, more than 20 years later, seeing the same story again felt disappointing. In short: The FDE concept they described and the example they presented did not seem to match.

This is not only a technical issue. It is also a question of people’s livelihoods. Many COBOL engineers are now in their 50s or 60s. Many experienced Java engineers are also in their 40s or older. If the message is simply “we can replace their work,” that feels uncomfortable. Even if tens of thousands of files are automatically converted, can a company really deploy them without human review? Of course not. People still need to check the output carefully. That means the workload does not disappear completely. Development costs remain. In some cases, the total cost may even become higher. Moreover, in enterprise systems, the core business logic is often only a small part of the entire codebase. A large amount of code exists for exception handling, risk prevention, and defensive logic. Those defensive patterns change from era to era. Simply translating code from one language to another may not preserve the real operational meaning behind the system.

To me, real value should be about helping companies make more money, not merely helping them save money. Companies that only focus on cost reduction have limited ambition. Truly successful companies think about how to invest money to create something bigger. From that perspective, approaching large enterprises mainly with a cost-saving message felt somewhat narrow.

Part Two: Three Questions That Made Me Uneasy

The second session was also presented by people related to the FDE team. Again, they were all very kind, and the speaker was excellent. They introduced several practical examples of using Codex, such as connecting ChatGPT with PowerPoint to generate polished presentation slides, and using Codex as part of a business workflow to coordinate Agents, Skills, MCP, and APIs.

Then came the Q&A session. I asked three direct questions, and the answers made me feel uneasy.

Question One: What Is Your Strategic Roadmap?

I asked:

Today’s session mainly focused on Codex. However, in May, OpenAI also released two important products globally: GPT RealTime 2 and the OpenAI Agent SDK. What are your plans to promote these in Japan this year? Or are you mainly focusing on Codex?

The answer was somewhat vague. The speaker said that they were certainly studying and discussing these topics internally, but the direction would depend on what partners and customers wanted. For this year, Codex seems to be receiving the most focus.

That answer made me feel concerned. In a large company, I would usually expect a clear roadmap, such as:

• What will be promoted in this phase?
• What will come next?
• How will Codex and Agent SDK be positioned differently?
• When would GPT RealTime 2 become the right entry point?

I expected that kind of strategic explanation, but I did not hear it.

I also felt some doubt about the FDE role itself. My ideal image of an FDE is someone who combines the strengths of a salesperson, engineer, and architect. Such a person should be able to identify the real problem behind a customer’s words, redesign the workflow, and use AI to accelerate implementation.

But what I saw felt closer to a traditional structure: Sales is sales. Engineering is engineering. That may be common in Japan, but the whole point of FDE should be to move beyond traditional Japanese-style IT delivery.

Question Two: Have You Considered Server Load and Reliability?

In their examples, Codex was positioned almost like the core controller of a business system. Users interact with it directly, and it calls APIs, writes code, and coordinates other systems. So I asked:

Do you have any plans for Codex to work with local models or on-premise deployments?

The answer was: Codex currently works best with GPT-5.5 through the cloud, and local model integration is not part of the current best practice. That answer was not wrong. But my concern was different.

If every company sends all requests to OpenAI’s cloud servers, what happens when usage explodes? Can the infrastructure handle that level of enterprise dependency? If many users treat Codex like a general chatbot and send countless casual requests, could that overload the system and affect serious enterprise users?

What I hoped to hear was something closer to NVIDIA’s “AI factory” concept. For example, companies could build their own internal AI infrastructure. Codex could act as the interaction layer. Routine tasks could be handled locally, while more complex problems could be escalated to cloud-based GPT models. That would be safer and likely easier for conservative Japanese companies to accept.

However, I did not feel that this direction was currently part of their thinking.

Question Three: Should I Join You or Compete With You?

My final question was a little direct. I asked:

In the current FDE trend, do you think I should try to join OpenAI Japan, or should I build an FDE team at my current company and compete with you?

The reason I asked this question was that I wanted to see their ambition. If I were in their position, I would probably respond with energy: “Great question. Let’s talk. We would welcome people like you. Or perhaps we can explore a partnership and grow the market together.” To me, the most important thing would be to attract capable people and expand the ecosystem.

But their answer was quite neutral. Someone from the PR side stepped in and said something like: “Well, that is a rather personal question, but we should all work together to grow the market.” That was it. There was no clear or ambitious answer.

This made me feel even more uneasy. As a company that may aim for an IPO, I expected to feel more ambition, a stronger strategy, and a greater sense of urgency to capture the Japanese FDE market. But I did not.

Conclusion: Perhaps It Was Just a Small Event

Of course, it may be unfair to judge the entire company based on one event. This event was not hosted directly by OpenAI. It was organized by another event company. Perhaps the event itself was too small, and OpenAI Japan simply did not reveal its real strategy.

Still, this was a useful reminder for me. OpenAI Japan is not invincible. The FDE market is still immature, and its future direction still needs to be shaped by many companies and individuals. In fact, because large companies may not have the bandwidth to deeply cultivate every local market, I may have found an opportunity. Perhaps if I follow the theory and approach I believe in, I may not be at a disadvantage.

These are only my personal opinions. I do not intend to attack OpenAI. On the contrary, I genuinely like OpenAI. I am a loyal fan of GPT, Sora, and Codex. That is exactly why my expectations were so high. I even considered joining them. So when the reality felt less impressive than expected, the gap felt larger.

Finally, this article does not constitute investment advice.

I sincerely hope OpenAI continues to grow. But for me personally, I may begin to see it more as a powerful tool provider rather than a company I deeply admire. Other companies are catching up quickly. I worry that Codex may have made a strong start but may struggle to maintain its lead, just as OpenAI once lost some momentum in multimodal AI to Gemini and faced strong competition from Anthropic in coding.

For now, I am looking forward to Google Cloud Next Tokyo 2026 in July. To be honest, I may currently feel more drawn to Google’s approach. Their enterprise strategy appears clearer and more structured. Putting all data in the cloud also carries risks, of course. But at least I can feel their determination and planning when it comes to doing business.