The stamp at the top of this IT Home page is 2026/10/23 12:04:21. It retells a Nikkei Asia report from the 22nd. The Japanese company EpicAI plans to help Japanese carmakers use artificial intelligence to speed design reviews for new vehicles. The software is planned to reach several Japanese carmakers from the spring of 2027. The company says an earlier trial drew positive comments from customers. Positive comments are not scores. This page prints no timed comparison.

[1]
A charcoal clamp, one muted blue screw
A charcoal clamp with one muted blue screw, one darker screw, and a little oxide red in the shadow under the blue screw., AI-generated illustration, not a news photograph

The hours come from chief executive 横山敬一. Reviewing the design of a single part can take about 200 hours. EpicAI hopes its software will cut that work from about 200 hours to a few hours. Hope, and a few hours, are not a timed result already finished. He also gives an example: cutting the design process in half is far from easy. Cutting in half is the hard problem in that example, not a ratio the software has already achieved. Two hundred hours against a few hours, and cutting in half, are not one result.

[1]

The rollout is not company-wide either. The software will first be used in some departments of several Japanese carmakers, and whether to expand still needs further discussion. Some departments, plus further discussion, mean spring 2027 is not a date for switching an entire company. EpicAI was founded in 2023, is based in Tokyo, and has about 50 employees. Its work includes artificial-intelligence consulting and custom systems. The customers named on the page include Denso and Panasonic, mentioned as a Tesla supplier. Those are existing customers. They are not a list of vehicle programs whose reviews have already been shortened. The page also says the company grew out of the school it names 东京科学大学. 横山敬一 previously worked in the lab of 松尾丰. The page says 松尾丰 has long taken part in Japan's artificial-intelligence policy and chairs the government's artificial-intelligence strategy expert group. That is a biography, not a count of hours already saved. 横山敬一 also says many vehicle recalls come not from manufacturing quality but from failure risks the design did not foresee. That is his judgment. The page attaches no recall list.

[1]

The page also says the software uses several generative models that split the work, looking at specifications, risks, and failure cases, then checking one another. That is a method. This story does not repeat the split and does not turn it into a configuration. What remains is the judgment: about 200 hours is the company's account of reviewing one part today, a few hours is the time it hopes to reach, spring 2027 starts with some departments, and cutting the process in half is only the chief executive's example of how hard the job is.

[1]