[Workshop] Hosting an "AI Business Development Workshop" in collaboration with Plus Co., Ltd. and SAMURAI ARCHITECTS Co., Ltd.

　On April 27, 2026, Plus Co., Ltd., SAMURAI ARCHITECTS Co., Ltd., and SIGQ Co., Ltd. jointly hosted an “AI Business Development Seminar.” At the seminar, our CEO, Takaaki Kanetsuki, took the stage to discuss the current state and future potential of AI from the perspectives of business development and physical AI.In this article, we provide an overview of the presentation, which is also relevant to SIGQ’s mission of “Designing Reliability.”

A Scale of Business in the AI Era Rivaling the National Budget

At the outset, Kanetsuki began by explaining the technical fundamentals of AI.

When you use ChatGPT, Claude, Gemini, and similar tools on a daily basis, it’s easy to think of AI and LLMs (large language models) as the same thing, but LLMs are really just one type of AI. Deep learning and machine learning fall under the broader umbrella of AI, and within that field, LLMs have been gaining particular prominence in recent years.

Why are LLMs able to provide natural responses to human questions? LLMs learn contextual word connections and patterns from the vast amount of data available in the world, such as books and web content.When text is inputted, the system calculates the probability of which word would follow next to sound natural from a vast pool of candidates, and outputs the word that seems most likely (with the highest probability). By repeating this process, it generates coherent responses. Services utilizing this mechanism have already been incorporated into many business operations, such as summarizing contracts, translation, drafting emails, and creating documents. Modern AI is supported by the processing of massive amounts of data.

Next, Kanetsuki discussed the scale of business growth in the AI era. Using a quiz format, he asked the participants how long it took Uber to become profitable and how much it cost.The answer was 14 years and approximately 4.5 trillion yen. To grow a business to the point where it could be considered social infrastructure, it required that much time and the resolve to absorb massive losses. On the other hand, Anthropic, the company developing Claude—a rival to ChatGPT—is said to have achieved annual revenue of 4.5 trillion yen in just five years since its founding. Furthermore, just before the lecture, news broke that Google had made an additional investment of 6.4 trillion yen in Anthropic.Today, an amount equivalent to 70–80% of Japan’s annual defense budget is being poured into a single startup. Furthermore, Kanetsuki conveyed the impact of an era in which the advent of AI is rewriting the very growth curve of business, while sharing the reality of the talent landscape in Silicon Valley—gained through his extensive overseas experience, including interactions with international clients and the tech community.

The Potential of Physical AI Expanding into the Real World

　Furthermore, the scope of AI is expanding beyond the digital realm into the real world. In light of this trend, Kanetsuki presented the latest case studies to illustrate how physical AI is improving both accuracy and practicality.

The first is the "AGIBOT G2," a robot for the manufacturing industry developed by a Shanghai-based company. It is designed for precision tasks—such as assembling and repairing tablet devices—where even the slightest misalignment can lead to catastrophic failure, and it performs its work while adjusting its position with millimeter-level accuracy. It is also equipped with the ability to navigate autonomously within the factory as needed.

Its most notable feature is its ability to learn from its own mistakes. Sometimes it receives guidance from humans, and other times it analyzes the causes and makes corrections on its own; by accumulating these experiences, it continuously improves its accuracy. Kanetsuki notes that its practical application is gaining traction even in precision equipment manufacturing facilities—where AI implementation was previously considered difficult—and that it is viewed as a promising solution to the labor shortages resulting from Japan’s declining population.

　The second is autonomous driving. While research into autonomous driving began about 15 to 20 years ago, the technology is nearing the stage of practical implementation overseas. In San Francisco, driverless taxis are already operating on city streets; simply enter your destination in the app, and they’ll take you there. He added that although data collection for autonomous vehicles has begun in Japan as well, there are still many challenges to overcome in complex environments such as Japan’s narrow residential streets.

　The third is "Ace," a table tennis robot developed by Sony. It uses a camera to detect where the ball is coming from and what kind of spin it has, then calculates the optimal angle and spin of the racket in less than a second to return the shot.Kanetsuki highlighted Ace’s advanced capabilities by recounting an anecdote in which it won every match against professional table tennis players. The fact that it has mastered a sport like table tennis—which requires continuous decision-making and movement in rapid succession—symbolizes the broad scope of application for physical AI.

The evolution of physical AI demonstrates that AI is no longer merely a tool for generating text or images, but is increasingly becoming an entity capable of making decisions and taking action in the physical world. That is precisely why the question of what happens when it behaves incorrectly is becoming increasingly critical for both business and society.

Designing Trustworthiness in a Society Where AI Is Becoming Ubiquitous

　During the Q&A session following the presentation, a participant asked a thought-provoking question: Since AI makes decisions based on probability, errors are inevitable. While humans can review and correct errors in documents, how can we ensure accountability when it comes to physical AI systems that involve life-and-death situations?

　Kanetsuki’s response to this is to incorporate the engineering concept of “fail-safe” into physical AI as well. Fail-safe is a design philosophy whereby, if an incident such as a malfunction or operator error occurs within a system, the system is designed to always operate on the side of safety rather than continuing to operate in a dangerous manner. For example, machinery on a manufacturing floor stops operating when it detects an abnormality.Similarly, Kanetsuki argues that physical AI must be designed to transition to safe operation when an incident occurs. In fact, in the field of autonomous driving, safety-focused controls—such as prioritizing braking when danger is detected—are already being implemented. To trust AI’s decisions, it is crucial not to assume that it will always be perfect, but rather to ensure that it is equipped with mechanisms to guarantee safety even in situations where it cannot make appropriate judgments.

In fact, the concept of fail-safe is deeply rooted in the insights SIGQ has cultivated over the years regarding incident response and the continuous improvement of reliability.

At the core of SIGQ’s business lies “design for reliability.” Kanetsuki himself has spent nearly a decade building his career as an engineer focused on big data processing and SRE (Site Reliability Engineering), while also conducting research on databases and large-scale distributed systems. This expertise is applied to our professional services that support incident response and to the development of “Incident Lake,” an autonomous AI agent.SIGQ’s business domain lies at the intersection of system reliability and the massive amounts of data essential to AI—these two areas of expertise.

Now that AI has entered the real world and begun to impact human lives and massive business operations, reliability engineering is evolving from a mere technical challenge into a cornerstone of business continuity.Know-how regarding the stable operation and rapid recovery of web services will merge with the fail-safe concepts cultivated by engineering and manufacturing, culminating in reliability design methodologies to realize safe physical AI. This will become indispensable as AI continues to evolve.

What Are the "Wrong Ways" and "Right Ways" to Use AI?

So, how will the role of humans change in the era of physical AI? Kanetsuki began by outlining the differences between the nature of AI and that of humans.

Where AI truly excels is in its ability to perform specific tasks 24 hours a day, 365 days a year. For example, the ability to quietly and efficiently translate 10,000 contracts by the next morning without a single complaint is something only AI can do.

On the other hand, one of the advantages of humans that is often overlooked is their energy efficiency. Running AI systems requires enormous amounts of water and electricity, and energy is consumed right from the semiconductor manufacturing stage. Humans, however, can sustain their activities with just food and a moderate amount of water.

There are many other things that only humans can do. For example, building relationships, making important decisions that carry significant responsibility, and exercising sophisticated judgment in specialized fields.Activities such as experts exchanging information at academic conferences or deepening relationships with clients over drinks are uniquely human. Furthermore, in fields that require a high level of expertise and critical decision-making—such as life-saving surgeries—even if physical AI provides increasing support, humans will continue to be the ones responsible for the final actions and judgments.

Based on these differences in characteristics, Kanetsuki proposes the following guidelines for the proper use of AI.

One common mistake in using AI is handing over tasks to it in areas where you cannot review or evaluate the results yourself. Even if the AI responds with “Done,” if you cannot judge the accuracy of the content, the results will be released to the public without you realizing there are errors. This is a classic example of why AI is often deemed “useless.”

The correct approach is to interact with AI in the same way you would when delegating a task to a subordinate. Just as a manager doesn’t simply bring documents received from a subordinate to an executive meeting without reviewing them first—but instead reads through them, makes corrections, and then submits them—Kanetsuki says the same mindset is necessary when dealing with AI. To fully unlock the value of AI, it is essential to delegate tasks that you can review and take responsibility for, carefully evaluate the output, and make adjustments as needed.

Kanetsuki also recommends using AI to reinforce and review the materials you’ve created. If you ask it, “Is this wording appropriate?” regarding materials for management meetings or important documents, it will instantly point out any contradictions or redundant sections. Precisely because you have a deep understanding of the subject matter, you can effectively incorporate the AI’s feedback.

Summary

　With approximately 100 participants attending both in person and online, the seminar received high marks in the post-event survey. Ninety-two percent of participants reported that they “learned a great deal,” and 98% said they “would definitely like to attend again,” demonstrating the high demand for opportunities to learn about the latest trends in AI through real-world examples.

　In an era where AI is expanding its role into the real world, creating mechanisms that transform behavior governed by probability into something “reliable” is essential for translating AI into business value. SIGQ will continue to support companies in their use of AI through reliability engineering.