Read the story in Japanese

KANAZAWA, Japan –Tetsuya Kitanishi stands outside a metal cubicle the size of a sauna shed and taps its display panel. An avatar of a woman in a bright yellow jumpsuit springs up and asks him, in Japanese, to select the file for the precision part he wants to make. It then analyzes the file and says, “Would you like to set up the machining process?” 

Meet KAYA, the conversational AI interface for TTMC Origin. This machining center is being prototyped by ARUM Inc., which builds fully automated manufacturing systems in Kanazawa, Japan. 

Powered by ARUM’s in-house software ARUMCODE and built using Azure AI Speech and Azure OpenAI in Microsoft Foundry, KAYA guides workers using the machine to make precision parts, instructing them step-by-step in natural language to change a drill bit or reposition the item being cut. 

The idea is to build a CNC (computer-numerical control) machining center that can be operated by junior workers and not just skilled experts, a dwindling breed in Japan. 

According to Kitanishi, the lead software developer for TTMC Origin, incorporating generative AI into the machine to voice these intricate manual steps forms a key part of the solution. KAYA currently runs on GPT-5, OpenAI’s large language model. 

This is among the newest projects for machining solutions designer ARUM, which is using AI to help Japan’s globally ranked precision part makers overcome a labor crunch.

This shortage, described by the Ministry of Health, Labor and Welfare as “long-lasting and persistent” since the 2010s, has meant demand for workers in manufacturing continues to outstrip supply.

“To address the sharp decline in skilled machinists in Japan and abroad, we enhanced our automation technologies and began developing ARUMCODE and TTMC,” said ARUM founder and CEO Takayuki Hirayama.

Japan’s precision manufacturing sector is worth about $15 billion at current exchange rates, estimates Marc Einstein, research director at technology market research firm Counterpoint Research. 

“Japan has maintained its global leadership in this area as it holds about a 60% market share in some very specialized and important areas like precision equipment for semiconductors, robotics and optics,” said Einstein. “These will become more important with the rise of physical AI.”

Translating craftsmanship to code

Takaaki Sakashita, general manager of ARUM’s software development team, acts as a bridge between old-school machinists and software programmers. 

When he began his career designing precision parts for medical devices over two decades ago, he learnt the craft the traditional way. 

This meant observing and absorbing how his seniors produced digital drawings, chose materials and specified which tools a machine should use in what sequence to shape a part. It took seven years before he was experienced enough to make these decisions on his own.

Today, he too is passing on that tacit knowledge, but via code. 

“My job is translating the craftsman’s intuition for the programmers, who need to put this expertise into numerical form,” said Sakashita. “We are creating the AI brain.”

These programmers, with Sakashita’s input, built a vast database of part materials and shapes, cutting patterns and tools, then trained ARUMCODE via a graph neural network on how these myriad items relate to each other. 

This enables the manufacturing AI to analyze shapes and automate the time-consuming process of turning computer-aided design (CAD) drawings into computer-aided manufacturing (CAM) instructions for the machines. 

According to ARUM, it used to take a skilled machinist over an hour to create a machine program for making an aircraft wing rib the size of a mobile phone. With ARUMCODE, it only takes four minutes.

This AI software also powers the TTMC Type F, ARUM’s first machining center that physically executes those instructions with minimal human intervention. 

Both ARUMCODE and TTMC run on Microsoft’s Azure cloud platform.

“If we had ARUMCODE and TTMC in my old job, we could have designed and made prototypes of dialysis machine parts in three weeks. Instead, it used to take up to six months,” noted Sakashita.

For producers of so-called “high-mix, low-volume” precision parts – which are very customized with a higher per unit cost – shorter production cycles and lower costs can make all the difference between a profit or a loss. AI solutions like ARUMCODE and TTMC can also keep factories running in the face of Japan’s shortage of skilled labor. 

“The job openings-to-applicants ratio in the manufacturing sector is 1.67, so it’s fair to say there will be a shortage of several hundred thousand positions in the next few years,” said Einstein. “The average age of a mechanical engineer in Japan is already over 50.”

These were all issues CEO Hirayama set out to solve when he decided to pivot towards automation and AI back in 2008.

At the time, ARUM was a subcontractor with 20 employees, designing and producing parts for the auto and semiconductor industries. 

“The work we were doing was very difficult, and the profits were very limited,” he said. “Industry-wide, we were seeing a major decline in the number of skilled workers who can work in metal processing.”

When the global financial crisis hit that year, many of ARUM’s peers went bankrupt. “So, we thought, there will be demand for our technology if we build it,” he said.

It took another 12 years before ARUMCODE was ready to ship. 

The most challenging task for the software developers was turning some 4 million cutting conditions into one algorithmic formula, said Satoshi Murakami, general manager of the cloud system engineering department.

“That was very, very hard. There was so much data that our Excel spreadsheets would freeze,” he recalled.

Now the team uses GitHub Copilot to code more efficiently.

ARUMCODE became commercially available in 2021. After which ARUM, an alumnus of the Microsoft for Startups program, partnered with Sugino Machine Co to build the TTMC Type F, launching it in May 2025. 

By combining both AI-powered solutions, ARUM has automated metal processing’s entire 12-step production process from drawing to finished part and made them operable by less-skilled workers.

 

New direction, more plans 

 ARUM, now with 40 employees, has also created a new business direction and revenue stream for itself in the process.

According to CEO Hirayama, ARUM has sold 40 TTMCs so far at 330 million yen ($2.1 million) each. It also has over 200 manufacturers in Japan plugging ARUMCODE into their own machining centers via their subscription to ARUM Factory 365.

“If we were to compare the company now and back then, profits have increased eight- to ten-fold,” said Hirayama. “The company name has also become known worldwide, unlike when we were a subcontractor. We now consider ourselves a fabless company, focusing on design and development.”

Adding AI upgrades to its technology, like KAYA for TTMC Origin and GitHub Copilot to write code, is just one of many plans.

ARUM also plans to build a procurement network of linked TTMCs across Japan so that if production is disrupted in one area, for example due to an earthquake, it can be backed up by machining centers in another location.

“This network can be controlled in a consolidated manner on Microsoft Azure,” said Hirayama. He’s confident that security features in Microsoft’s cloud platform will safeguard communication between machines and manufacturers’ design data.

Azure’s scalability and global reach will also support ARUM’s plans to eventually export TTMCs to countries like the US, South Korea and India, said Murakami.

Without AI, Hirayama thinks the company, which he founded at his kitchen table in 2006, could have carried on with just its precision parts business, but its growth potential would have been limited.

“We would survive, but without any big dreams,” he said.

Top image: Software developer Tetsuya Kitanishi with KAYA, the conversational AI interface being prototyped by ARUM Inc for its machining centers. Photo by Noriko Hayashi for Microsoft. 

Lim Ai Leen reports on AI for Microsoft Source, focusing on how it’s improving lives in Asia. Ai Leen was formerly associate foreign editor at The Straits Times in Singapore and still pens an occasional weekend column. Contact her on LinkedIn.