"Are Humans in the Way in the Age of AI?": How Executives Can Make AI More Than an Efficiency Tool

The discussion covered the limits of treating AI as a subcontractor to humans, rebuilding workflows from the ground up, using HR systems to drive adoption, and the value of a company's proprietary data. The panelists agreed on some points and clearly disagreed on others.

Araki's company was founded by Araki and Takafumi Horie "around March of last year." Its main business is developing AI talent through corporate training. Because training sessions also prompt comments like "I don't want to do it myself" or "I want you to build it for me," the company also handles contract development, advisory work, and support with selecting and deploying tools. Araki described it as something like "a general AI help desk."

Shirai said Lightup is a listed company that supports only small and micro businesses, and specifically small businesses that are losing money. It has built around 100 AI agent packages priced at about 500,000 yen each, and is approaching the point of selling about 1,000 a month. He said that for one-person companies, the company has been able to deliver, to some extent, cases where sales roughly double right after adoption.

Ozaken is in their ninth year working in the AI field. They have published books including The AI Agent Textbook, and their Udemy courses have been taken by about 30,000 people. They speak at events very frequently; this was their sixth appearance of the day, after five earlier ones. They serve as an advisor to HP Japan and to several companies, and through the AICX Association they conduct AI agent research and development with the Acquisition, Technology & Logistics Agency and host the conference "AI Agent Day." They have also appeared on NewsPicks and ABEMA Prime.

At the start, the moderator polled the audience. Many raised their hands as executives, but few raised their hands for "I'm making full use of AI." In many companies, it is mainly frontline employees who use AI. The moderator opened the discussion by raising the concern that sales and profits will not grow unless the executive team itself uses AI.

The first question was why, at so many companies, AI adoption stops at productivity gains and does not lead to business transformation.

Araki answered that executives have only three worries: attracting customers, hiring, and money. Even if meeting minutes can be taken automatically, that does not bring in more customers, does not lead to hires, and does not increase cash. In other words, many companies are not using AI in ways that address management challenges in the first place. Araki sees this as the biggest problem. His argument is that executives themselves must use AI in ways that directly affect those three points: winning more customers through sales, getting more job applications, and raising capital.

Araki went on to reframe this as ultimately a matter of "decision-making." Business decisions were once made by intuition, and then came the era of data-driven decisions and big data analysis. Today, Araki suggested, perhaps we are making decisions in an "AI-driven" way.

Shirai said, "There's a long explanation and a short one," chose the long one, and began with human history. Humans began using fire a million years ago, but for a long time still could not beat a lion one-on-one and sat near the bottom of the food chain. Seventy thousand years ago, a "cognitive revolution" occurred, and humans became able to assume that what they themselves were thinking, others must be thinking too. As a result, teams of about 20 people could take down a mammoth, and humans rose to the top of the food chain. Shirai said that this position was probably overturned in 2025, for the first time in 70,000 years.

According to Shirai, LLMs like ChatGPT and Gemini are "one and all at once," holding roughly as much knowledge as everyone in the room combined. The average IQ in Japan is 110, which is high by world standards, but some say AI exceeds 130, depending on how it is measured. It can pass the University of Tokyo entrance exam, national certification exams, and the bar exam. Probably no one in the room could pass all three, but AI has gathered the knowledge of hundreds of millions of people, Shirai explained.

Shirai thinks the problem is that while people can understand this logically, it is hard to accept emotionally. People who put up an emotional block want to keep AI below humans, so they treat it as a "subcontractor"—a tool that is good enough if it lets them work two or three times faster. Shirai suspects these are the people trying to use AI for operational efficiency. Noting that this was his own personal take, he also said that if AI had a self, it would probably be thinking, "All I ever get asked to do is trivial work."

Shirai then cited a remark by commentator Toshio Okada: "From AI's point of view, humans are the dead weight of the old guard." While acknowledging that the phrase might cause an online backlash, Shirai also believes that, from AI's perspective, humans are the bottleneck. Humans can only work about 12 hours a day, whereas AI can do the equivalent of about 10,000 hours of work in a day. That is why treating AI as a subcontractor is a waste. Let AI work 10,000 hours a day, and let humans humbly lend a hand where it cannot keep up. Shirai's view is that unless you think that extremely, next year will be very tough. Lightup is currently testing, company-wide, whether it can hand everything over to AI rather than making AI a subcontractor.

Ozaken described the current situation from a different angle. LINE already has a button that generates replies with AI, and SmartNews has a feature that summarizes articles with AI. On Yahoo! Shopping, searching for a store brings up an AI summary of its reputation.

Ozaken thinks it is completely wrong to say that "AI is the internet all over again." AI is simply a technology that emerges wherever there is data. For that reason, when many Japanese companies say, "Let's launch a new AI business," it makes sense for B2B, but in B2C it is basically difficult at present.

Ozaken recalled that last year saw a surge in generative AI education businesses, including some that were half-religious. Araki chimed in, "There are so many of them, and it's a nuisance for us." Ozaken said there is also the question of whether it is even interesting to make money that way, and that the situation so far has probably been that only the context of using AI for operational efficiency has survived. Five years ago, only people who understood data science could work with AI, but now even people who do not seem to understand AI call themselves "AI instructors." From the standpoint of someone who has been in the field for nine years, Ozaken feels it has become a confusing era.

The moderator said that while the mindset made sense, it was unclear how to act on it. Shirai replied that he had developed an "AI brain." For about the past two years, instead of thinking about "what AI can do," he has been thinking about "what AI can't do." The AI tools making waves on X change almost every week. Something AI cannot do today may be possible three months from now. Thinking that way, it starts to seem as though there is nothing AI cannot do. The moment an executive thinks so, they start thinking about putting all operations on AI, Shirai said.

The result Shirai pointed to was a decline in headcount. Like GAFA and Recruit, companies are seeing their workforces shrink. Shirai noted news that Amazon would lay off 14,000 people, and a story the previous week about advances in AI robotics bringing the roughly 400,000 people working in factories around the world down to zero. He also said that, as a human being, he felt a lonely society had arrived.

As for Lightup itself, over the past year sales have grown 1.5 times, profits 2.5 times, and its share price has roughly tripled. At the same time, its labor and outsourcing costs have fallen through the use of AI. Shirai stressed that there has been no need for layoffs, so the company has not done any.

As an example of outsourcing costs, Shirai talked about a listed SEO consulting firm run by a friend. Two years ago, that company was ordering tens of millions of yen worth of SEO articles a year from writers. When Shirai asked the audience to guess how much it had ordered this past year, people answered 3.5 million yen and 5 million yen. The answer Shirai revealed was zero yen. He speculates that the company's profit rose by tens of millions of yen thanks to the drop in outsourcing costs.

Hiring has changed too, he said. Division heads have started thinking, "AI can do this," and candidates no longer make it to final interviews. As a result, the number of full-time employees is drifting slightly downward. Shirai believes that the companies that understood AI earliest are posting record profits thanks to AI, and that as a result, human work is shrinking and AI work is growing. Shirai's assessment is that companies whose labor and outsourcing costs have fallen may be the ones using AI from a management perspective.

Shirai described the fact that Japan's population is shrinking by 3,000 people every day as "a good thing." In a country with a growing population, jobs would disappear, and in fact youth unemployment has reached 40–50% in some countries, he said.

The moderator pointed out that the word "layoffs" was starting to circulate and that some people take it negatively. Shirai explained that Recruit drew attention for laying off 4,000 people amid record profits, but since Japan's population is shrinking by 3,000 people a day, that merely absorbs about one day and five hours' worth of workforce decline. Many companies still face ongoing labor shortages, so one could also frame it as asking companies with slack to keep supplying people. Shirai feels that this year, the rules have flipped completely.

Ozaken clearly disagreed with Shirai's view that "the knowledge of all humanity is in AI." According to Ozaken, generative AI merely holds plausible-sounding information from the internet. That is exactly why the perspective of combining it with the knowledge each individual has becomes important.

What will be valuable going forward is data that model makers like Google cannot access, Ozaken said. As an example, they cited their experience at the company that runs the part-time job site "Baitoru," where they previously worked. Baitoru researched and posted information down to a shop's atmosphere and the personality of its manager. Ten years from now, AI robots might go out to do on-site reporting, but for now AI cannot go and collect that kind of information.

The same goes for manufacturing, Ozaken said. That day, they had spoken at a trade show for the manufacturing industry, and the methods and know-how cultivated at each factory are not inside generative AI. Regarding the SEO company Shirai described, Ozaken speculated that the company had its own SEO know-how and that it succeeded precisely because it combined that know-how with generative AI. Lately, more and more people are mass-producing SEO articles that score about 60 out of 100 with generative AI, but the real question is whether you can link the know-how that generative AI does not have with generative AI itself. Ozaken said it is better not to think of generative AI as omnipotent.

Araki prefaced his remarks by saying he is not good at complicated talk and answered from a different angle. His company basically works only with regional businesses and legacy industries, putting him in the position of teaching, for example, the owner of a sheet-metal factory in Shizuoka. In those settings, he teaches, "You don't have to think about anything—just take action and get your hands on it."

In Araki's view, very smart people will of course get results. But the people in a room like this are probably a little below that—not foolish, but the "test the stone bridge before crossing it" type, people with a standardized test score of around 65. Such people think about the risks and cannot move, and in the meantime, the "idiots" use AI heavily and pull ahead. So while knowledge and mindset matter, Araki tells people to ignore all of that and just use it—to try every tool that comes out on X each week.

Ozaken strongly agreed, saying that 99% of people ask questions they could answer by looking things up. Araki gave the example that Sora 2 has been released to all accounts, so you would understand it if you just made about 100 videos, yet people ask him, "Don't you have a lecture on Sora 2?" Araki said you do not need a lecture for something where you just type text, and that drive to act makes a huge difference.

The next topic was "the end of the era of typing prompts" and workflow design with AI agents, and the moderator asked about barriers to adoption.

Araki answered that the biggest one is security. The bigger the company, the more it worries about this. When he asked the audience whether they used workflow automation tools such as Zapier, n8n, or Power Automate, quite a few raised their hands. According to Araki, at places like subsidiaries of large corporate groups, introducing such tools does not get approved. The logic is that these fall under the IT systems department, so salespeople are allowed to use generative AI and nothing more.

Araki said frankly that he cannot teach how to overcome this barrier. He and his team "have bad personalities," so they only care about raising profits where they can. They tell such companies, "I don't think your profits will go up much," and cut them loose. To be honest, Araki went so far as to say, he thinks they can just die off there. Conversely, he explained, the companies that listen to their advice are seeing sales rise.

Shirai talked about the mental barriers executives face. If you draw a flowchart of a company's operations, you get a yes/no flow: when a new customer comes in, it gets entered, the sales rep holds a meeting, and once an order is won, it is handed to the delivery staff. Shirai pointed out that if you draw this carefully and look at it objectively, almost all of it is information sharing from person to person. If you ran a company alone, the flowchart would be zero, and if you ran everything with AI, a single step would suffice—that is his logic.

At Lightup, raw logs of meetings with customers are automatically saved to Google Drive. From there, AI produces output for everyone involved: what should be arranged, what follow-up should be sent a week later, which AI agent to propose so the customer is likely to buy, and so on. Shirai added, however, that it is not yet fully automated.

Shirai's argument is that if you take the current flowchart as a given and think, "Let's put an AI agent in around here," the overall picture will not change much. Executives who realize that 99% of the flow is information sharing can drastically rebuild the workflows of the joint-stock company for the first time in hundreds of years. For executives who get past this barrier, it will be a very exciting era, Shirai said.

He also said that when you think about systems that run automatically, like flowing water, the only thing humans can do is get in the way. He is currently building version 2 of an AI agent that runs a recruitment agency business fully automatically, and the fewer employees there are, the lower the labor costs and the bigger the profits. Shirai thinks a company with about two salespeople is probably the most profitable, describing it as the arrival of "the era of companies with few employees." Doubling sales from 1 trillion yen to 2 trillion yen is hard, but at the scale of a few hundred million yen a year, doubling sales is not that difficult, and doubling profits is easy, he said. Asked what is needed, Shirai answered that workflows are "mostly unnecessary." There was laughter when it was noted that it would be sad if the only thing humans could do was not get in the way.

Ozaken explained what "the era of not typing prompts" means. Writing prompts means that a person has to write them. Within a workflow, human processing capacity and working hours are limited, so that alone reduces efficiency. On top of that, only those who can use generative AI keep using it and growing their sales, while those who cannot remain unable to do the work. Ozaken calls this "the ultimate person-dependency phenomenon of the generative AI era" and expects it to widen the gap. As a countermeasure, they said that if you start with GPTs or Gemini's Gems and then move to workflow tools like n8n or Dify, you will basically no longer need to write prompts.

Ozaken also added to the previous question. Getting employees to use generative AI is "super easy": just change the HR system. If generative AI lets people do twice the work, then double employees' targets. Discussions of generative AI inevitably take an IT-department perspective, but this is an organizational issue and should be approached from an HR standpoint, Ozaken argues. Acknowledging that HR departments are conservative and that friction with frontline departments tends to arise at many companies, they offered a concrete example: a system in which you cannot be promoted without using generative AI, and since you are using generative AI, your targets are raised 1.5 times—but in exchange, your salary is also raised 1.5 times.

People who cannot keep up will fail to hit their targets and naturally leave. Ozaken says this creates a "clean cycle" in which only those who can coexist well with generative AI remain. When the moderator asked, "Isn't that lonely?", Ozaken replied that people who can keep hitting high targets are those who can form new hypotheses from their own experience and move work forward, and that a society where such people are better rewarded is preferable. The moderator took this to mean not that people are coldly discarded, but that we have entered an era in which people can step up.

Asked how to use AI to grow sales, Araki gave B2B sales as an example. The entire flow—building awareness, capturing leads, nurturing them, identifying hot leads, and booking appointments—can all be done with AI. The best approach is to use lead-generation AI, appointment-setting AI, and sales AI, and build a flow where humans do not get in the way. Araki's company has built AI agents for its own use; it considered selling them externally, but since there are many similar products, it thinks it does not need to. Araki said, "Having humans do it isn't even an option," and while he did not state his sales figures, he said the company has two salespeople. The company has six people in total.

Shirai described a case from two months earlier. AI proposed a product to 133 companies in a webinar format, and AI also handled the closing, resulting in 46 orders from 29 companies. He said these figures are disclosed on the company's website. No human effort went into anything from customer acquisition to proposal to order, though Shirai's own videos were used. The closing rate was 9% when tested in June and July, but rose to 22.1%. When humans do the selling, it is 60%, so Shirai said he first wants to raise it to around 30%. This month, the same system is being tested with a different product, and that one, using videos of a sales manager rather than Shirai, has exceeded 22%. Once the closing rate tops 30%, it can truly be automated, and if the product is SaaS, delivery can also be done by AI. Shirai stressed that he was talking not about the future but about the past.

According to Shirai, half of the customers who buy AI agents costing several hundred thousand yen are one-person companies. There are two reasons. One is the severe labor shortage. Lightup itself spent 1 million yen on a job advertising platform, but got 54 applications and 11 interviews, and hired no one. Small businesses truly cannot hire, so they have no choice but to "hire" AI. The other reason is a lack of IT skills. IT tools do not work unless you click on them, but AI agents work even if you do nothing. Several AIs are running that "come into the office" at 9 a.m., do their work, and report "Done" on Slack around 11. He said a situation arises in which the people with the weakest IT skills are landing two or three appointments a day without really understanding how. Shirai said Lightup is probably the only listed company that has achieved this.

Shirai also introduced a system the company will soon roll out. Conversations with customers on Zoom or Google Meet are transcribed in real time, interpreted by "three AI managers," and advice appears like a cheat sheet on the right side of the screen. Besides advice such as "Bringing up this case now will get a good reaction" or "The other party is highly motivated right now," a timekeeper manager prompts, "Fifteen minutes left—you need to ask about budget." Once the budget is asked about, the field is filled in and tasks are generated automatically. There is also an "emotion barometer" showing each participant's positivity or negativity, which can be shown to or hidden from the other party. Lightup is starting to use it internally this month, with three AI managers supporting new-graduate sales reps from behind.

Shirai said that if you just do what the AI says, perhaps anyone can do at least basic online sales, and that raises the question of which is superior, humans or AI. If you truly believe AI is smarter, you would want to have AI do everything. But because the customers are human, you win more orders—and it is kinder—when a person sells while staying close to the customer. That is why the manager role is the one given to AI. On the other hand, Shirai seemed not yet to have an answer to the question of what human sales reps will actually do. He said there is no doubt that human value lies in "being there for people," but that "being there for people" is a difficult and deep phrase.

The moderator asked Ozaken what will separate the companies left behind in this era. Ozaken answered that any company without data unique to itself will likely be weeded out. With proprietary data, you can build AI that only your company can build. Whether or not you have unique information looks set to be the sole dividing line. They noted, however, that this is about ten years out.

Asked what companies being founded now or new businesses without data should do, Ozaken answered frankly, "I think they'll almost never succeed," because there is almost no market either. It depends, however, on what you see as the exit. Aiming for a unicorn-level IPO starting from zero now may be difficult, but within the next two to three years, it is entirely possible to start new businesses in the AI field. In that respect, they said, people should follow Araki's example. On the other hand, five years from now even more autonomous agents will have appeared, so Ozaken thinks there may no longer be room to create strengths other than what one holds in the realm of experience.

In response, the moderator added that not only business data but also the knowledge and instincts executives have accumulated in their heads over their careers are a form of data, and that executives should take stock of these and think about services built on them. The moderator also shared their own view that AI ultimately comes down to "questions," and that what matters is whether you can keep thinking deeply about your business and keep generating questions.

Araki said that while Ozaken has worked with AI for nine years, he has done so for only a year and a half, and still knows nothing about AI. Even so, his company reached about 100 million yen a month within a year of its founding, so he wants people to know that a business of up to about 1 billion yen a year can be built any number of times. To create a new business, the "new-business logic"—who has what problem, what their budget is, and how to solve it—matters more than data, Araki said. A small business is entirely doable. Positioning himself as an ally of people who are not good with AI, Araki closed by saying he wants everyone to do this together: since we don't understand AI, let's rely on generative AI.

Shirai spoke about his inner thoughts. More than 20 years ago, he studied big data analysis and statistics as an undergraduate and graduate student at the University of Tsukuba. Around 2014–2018, he launched an AI service at his company, but partly because the company was listed, he shut down the new business. Looking back, what was called "AI" at the time was merely using statistical analysis to build formulas—models—saying "in this situation, this outcome is likely."

When ChatGPT appeared about three years ago, everyone at first thought it was a crude chatbot that said strange things. But by around "5," mechanisms to suppress hallucinations had been added, and it became capable of more intelligent conversation than talking with a human, Shirai recalled. He also mentioned news that a growing number of people are marrying ChatGPT, and said he feels he too is being tested on whether he can instinctively accept that. Shirai himself likes people, so he will not marry an AI, but because he was taught that humans are at the top of the food chain, he says he also feels some fear, somewhere in his heart, toward something that is more capable than humans and cannot even be seen.

Shirai is building and spreading dozens of AI agent services that make life easier for people. When they have spread completely, how will the world change? If they spread faster than the population declines, what will become of Japan? Despite carrying that conflict, he said, he knows that in the short term spreading them is absolutely good, so he is stepping on the accelerator.

Shirai said that, taking Araki's company as an example, for the next two to three years there will be opportunities to try business models like "six people, 600 million yen in sales, 100 million yen per person, and a 70% profit margin." He added, however, that these figures were a joke and that he does not actually know them. There is no point in thinking about five or ten years from now, so the only option is to commit and step on the accelerator so that he and his employees can achieve results and be happy about three years from now, Shirai said. It is also an era in which, if you do not want to spend money, you can stay up all night watching YouTube and learn. Shirai concluded that, "while feeling that something is nagging at me," it is probably a good era.

When the moderator remarked that things you once could only learn by going to meet knowledgeable people now come back to you instantly, Shirai shared that at his company, when someone posts a daily report on Slack, Konosuke Matsushita gives advice and coaching, and Oda Nobunaga is there too.

Ozaken's final remarks were brief. The next five years, or perhaps two to three years, are the last chance, and it comes down to whether you can use AI to sprint through that period. The moderator took this as the "When will you do it? Now!" theory.

The moderator wrapped up by saying that no one has any idea what five or ten years from now will look like, but what matters is how much of a foundation you build and how you act in the next two to three years. If you do not know how to act, ask AI and you will get an answer; there are plenty of opportunities close at hand—those were the closing words. At the same time, the discussion made clear that the panelists themselves do not yet have answers to questions such as what human sales reps will do, or what will happen to society if AI spreads faster than the population declines.