The market is finally within reach. CEO Office's Fujii on what he saw from both the "selling" and "buying" sides of AI

AI doesn't reach people just because it can be built. Hikaru Fujii, Program Manager at Corpy's CEO Office (VP of the CEO Office), has witnessed this truth twice, from two different vantage points. The first was as the manufacturing GM at an AI platform company, traveling to factory floors across Japan on the "selling" side. The second was as Head of IT at a cosmetics manufacturer, leading DX from the "buying" side. The real reason that AI gets built but never used, he says, lies not in the technology but in decision-making and organizations.

Today at Corpy, what Fujii is working on is not selling a product, but creating the conditions for a market itself to take off. IT consulting, a trading house, an AI platform company, manufacturing, and now Corpy. We sat down with Fujii — who, at every turning point in his career, has chosen dynamism over stability — to talk about his philosophy of work and where Corpy stands today.

I've put myself where the world was about to change

——At the turning points in your career, you seem to have consistently chosen dynamism over stability. Where does that judgment come from?

I've simply stayed faithful to what felt "fun" and what I genuinely "wanted to do." "At each moment, put myself where the world is about to change" — that impulse has always been my compass.

I chose IT consulting first because, back in the 2000s, I believed "IT will change the world." I then moved to a major trading house because it was exactly the period when manufacturers were expanding production overseas, and I felt "IT could take me abroad." At the trading house, I handled IoT and AI as products — mostly bringing IoT solutions into manufacturers and exploring what could be done with the data we gathered.

——And from there, why an AI platform company?

The trading house I was at happened to be a user of that AI platform at the time. We built an attrition-prediction model at a group company, and it was surprisingly accurate. For example, people who logged exactly the same overtime hours every month — in other words, those for whom unpaid overtime had become routine — were far more likely to quit. Patterns like that started to emerge.

Right when I was thinking "there's something genuinely interesting in this product," that AI platform company reached out to me. I felt something fateful in the timing, and above all I had that gut sense that "AI will change the world." I moved without hesitation.

"Anyone can easily build AI" never reached the front line

——At the AI platform company, as the manufacturing GM, you proposed AI adoption to many companies. How often did you see "it can be built, but it isn't used"?

The product back then was sold under the banner of "anyone can easily build AI." If you had tabular data, you could build a model by drag-and-drop. In just tens of minutes, you could experience the "Wow" of "oh, something plausible came out." Many customers were delighted.

But that's as far as it went. Take a factory production line: you can build a model that predicts a product's defect rate from material blend ratios and processing temperatures. From the R&D side, you can see "these conditions tend to produce defects." But actually changing the manufacturing conditions is a call for the production floor, and the R&D model doesn't reach that far. There's a silo wall. Even with a steelmaker, you can build an interesting prediction model. But when it comes to using that prediction to run a massive plant's operations, the answer becomes "well, hold on, not so fast."

We ran a lot of PoCs. Everyone said "Wow." But cases where it was actually built into operations and used were, in manufacturing, almost nonexistent.

——The PoC moves forward, but it doesn't go beyond that. Where did you feel the essence of that structure lay?

At a certain point I realized that it won't progress as long as the people on the ground just say "hmm, interesting" and stop there. From there I shifted toward involving senior leadership, and started setting up roundtable-style sessions for executives. By sharing each other's cases and internal rollouts, the thinking was, they might reach the decision: "Right, let's change our own business processes too."

Change the business process, change the way work is done. Unless that decision can be made, AI never becomes usable. Once I realized that, I put my energy less into selling the product and more into moving decisions and getting it used continuously.

——You felt the weakness of decision-making on the customer side. In what situations did you think "this is something a seller's effort cannot overcome"?

In the end, making AI usable requires an enormous amount of energy — that is, people, resources, and money. But at the time, the number of companies thinking seriously about it was overwhelmingly small.

In many PoCs, for example, the person in charge had a primary job and AI was a side task. When their main work got busy, it stopped. If even the person directly responsible treated it that way, those above them were even more part-time about it. On top of that, by the nature of the technology, it's hard to guarantee a 100% result, and there were any number of factors that made decision-makers hesitate. I think AI itself was simply not yet properly understood in that era.

In a very small number of cases — at one major manufacturer — they stood up a dedicated AI team of about ten people and brought in outside experts as management to own the decisions. That company's adoption progressed several times faster than anywhere else.

That's when I thought, "Then I'll go inside a manufacturer myself, drive AI adoption from within, and actually practice corporate transformation."

What I only saw once I stood on the buying side

——At the cosmetics manufacturer, you led DX as Head of IT. Standing on the "buying" side, what did you see that you couldn't see as a seller?

When I was at the AI vendor, the culture was to sell with "there's something this amazing — how can you not buy it?" It was a purely product-out logic. But from the receiving side, AI adoption is almost never priority number one.

For example, "we just had a security incident, we have to deal with that first," or "this is no time to be running PoCs" — those kinds of circumstances exist in abundance on the real front line. The seller says "it's this good, why won't you use it," and the buyer says "we have bigger problems." The gap in urgency is enormous — and I only truly felt that once I stood on the buying side.

At that manufacturer, when I joined, it was nowhere near a conversation about AI. The security management systems weren't in place, PC and license management was inadequate, sales information was managed in personal Excel files... it was a front line of figuring out how to bring DX to a company like that. The first thing I did was introduce Salesforce — centralizing information. It was one "I know we should do AI, but not now" after another.

I think the users of that AI platform at the time were probably seeing the same view. For many of the customers who stopped replying or didn't renew their contracts, it likely wasn't that "AI's priority dropped," but that "there was always a more urgent problem to solve."

——We also heard you ran into the reality of "the problem before data — there simply isn't data."

I believed I understood what AI can and can't do, going back to my days at the AI vendor — including that it won't work unless the data is in place. But once you actually stand on the buying side, the sheer weight of the work of "getting the data in place" is far greater than I'd imagined as a seller. At least at that front line at that time, I honestly didn't get the feeling that AI was usable right away.

——What illusions do the selling side and the buying side tend to hold?

The seller's illusion is that "just install this product and you'll immediately master AI and succeed." I was like that too when I was on the selling side — I couldn't see the processes before and after the product, or the customer's own environment. Without being able to map out the path to success, I'd let myself believe "install it and things will change."

The buyer's illusion is that "a good product will change everything." In reality, to adopt AI, a mountain of work outside of AI appears. Change business processes, change how people are assigned. Those organizational-change costs always come bundled with AI adoption — and yet neither the resolve nor the preparation for them is there.

With a core-system overhaul, everyone knows "this is a big deal," so they prepare the structure and the budget seriously before they take it on. But with AI, that difficulty hasn't yet been made socially visible. It actually demands resolve on par with a core-system replacement, yet the buying side is utterly unprepared for it. That's the view I could see most clearly only after experiencing both sides.

I sensed the tide had turned

——Having experienced both the selling and the buying side of AI, what made you decide to return to an AI company?

One reason is the rise of generative AI. With ChatGPT and Claude appearing, the world's barrier to AI dropped all at once. The "various hurdles" I'd felt back in my AI-vendor days are visibly coming down. I felt this could move things up another level — that working on AI inside companies was finally about to get genuinely interesting.

The other is a personal sense of conviction about XAI (explainable AI). At the AI platform of my former job, the feature customers actually used most was the one that visualizes which features matter, and how much, when the model makes a prediction. What matters when predicting a defect rate from material blend ratios; what matters when predicting strength from steel material properties; what matters when predicting yield from production-line operating data — all of that becomes visible.

It's one of the functions of XAI, and on the manufacturing front line it was by far the best received. People would have that "ah, defects happen because this is like this" moment of understanding, and that led straight into quality improvement. AI is only providing the trigger for the insight; a person takes it from there, makes the judgment, and connects it to quality improvement. This collaboration between people and AI was the single most powerful use case. When Corpy reached out to me and I heard they were focusing on XAI and QAAI, I thought, "they're saying the same thing." If they're seriously betting here, it has to be interesting — because they're digging into the very point users care about most.

——Were there other points where you felt Corpy was "different"?

I got a message from Corpy via BizReach, thought "what is this company?", looked it up, and found that one of my new-graduate cohort was inside (laughs). That's what made me decide to hear them out.

Then I spoke with Yamamoto, our CEO, and as I listened to the challenges around Corpy's growth, I felt that my experience — having sold a strong product, and having stood on the user's side — might be useful. Internal buildout and process construction were things I'd done at the manufacturer too, so I figured I could help there as well. The timing to join also felt just right.

Better a pirate than join the navy. But I want to have both

——What does the role of CEO Office Program Manager actually involve?

In a word, "the do-anything person." Troubleshooting. Showing up where there's trouble and putting my hands to work where there aren't enough of them.

In the past week, for instance, I spent my time writing a proposal for participating in a national AI-related project. I also handle the account office for a major customer, and I'm advancing the internal ISMS buildout. I'm building, from scratch, the operational buildout of our core business-management system, and as the next stage, a mechanism to make management and managerial accounting visible.

——Where do you find it rewarding?

I have a real sense of taking on the things that someone has to do. Corpy right now is in a phase where it has to advance internal buildout toward an IPO, and there's a lot that has to be turned into systems. With the old, passive structure of "customers bring us work," you can't control the business performance of a listed company. The dashboard, the budget-versus-actual management — we have to design and run them ourselves. This part — building the company as a system — I find genuinely fun as a challenge.

That said, personally I'm not all that fond of rigid things. Steve Jobs said "it's better to be a pirate than to join the navy," and Corpy, until now, has in a sense been a pirate-like company. People did what they wanted with their own discretion and ideas, delighted customers, and succeeded. But that way of working has little reproducibility.

Now, toward the IPO, we're entering a phase of "let's properly build it out and become a navy." But becoming the navy itself would be a waste — both personally and for the company. I want to keep Corpy's pirate-ish culture while making it coexist with a disciplined organization. Keeping a place where engineers, especially, can keep performing with the same peace of mind as before is, in the long run, what I think matters most. It's precisely because we're in this phase that we can design it that way.

Someone who can say "is this really necessary?" to the unwritten rules

——From your perspective, what kind of person can thrive at Corpy?

Someone with a pirate-like spirit. Someone who can move on their own discretion, by their own judgment, I'd say.

And one more: someone who can properly voice their own opinion. Inside Corpy there are all sorts of unwritten rules — things that have become taken for granted because we've done them so long. Someone with the nerve to ask, of those, "is this really necessary?" A new person coming in and throwing that question is, I think, the very thing the company most needs going forward. Someone like that will, I believe, find an environment where they can work freely and enjoyably.

——Finally, a message for those who will work together with Corpy in the future.

AI right now keeps delivering impact — to politics, the economy, corporate activity, and even individuals' daily lives. Now that AI has become a tool anyone can use, the "right decision-makers" I was searching all over for ten years ago may no longer need to be searched for at all. We're entering an era where the person who experiences AI's value can decide on adoption right there.

When that happens, the next question asked is "can that AI really be trusted?" The XAI and QAAI that Corpy is working on are exactly the domain that answers that question. An atmosphere is forming in the market: "AI you can't explain, AI whose quality you can't guarantee — that's not really usable, is it."

The market has finally come within reach, right in front of us. This year is the decisive one, I believe. Precisely because the market is taking off, we have to assemble the team quickly and get the delivery structure in place. That's why there's meaning in joining now.

We spend the better part of our lives on work — so if you're going to work anyway, you'd want to spend that time on work and products that matter to the world. To those who feel that way, I'd love for you to knock on Corpy's door.