Toyota Motor Corporation
Toyota gives 800 powertrain engineers O-Beya, a team of AI agents that answers design questions
Toyota built O-Beya, a system of nine generative AI agents that answers powertrain engineers' questions on topics such as vibration, fuel consumption and regulations. About 800 engineers have had access since January 2024. Toyota reports usage but has not yet measured time saved.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
some 800
Powertrain engineers with access to O-Beya since January 20241
nine
AI agents in O-Beya at the time of the Microsoft feature1
hundreds of times
Frequency of use per month, per the project lead1
more than 100
Requests per month, per Toyota's Azure Cosmos DB blog post2
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
Cars are changing fast. Hybrids, electric vehicles, batteries and charging add more parts to design, while many senior engineers are retiring. Toyota's powertrain department, which designs the engine, transmission and drive systems, relies on specialists whose knowledge sits in past design reports and in their heads. Finding the right document used to take a long time.
What they built
O-Beya, named after a Toyota term meaning "big room", is a set of generative AI agents that act like a room of experts available at any time. Each agent covers one topic, such as vibration, fuel consumption or regulations. An engineer picks one or more agents, and the system combines their answers into one reply. The project lead says the system should eventually choose agents on its own.
How they did it
The system runs on Microsoft Azure and uses OpenAI's GPT-4o model. It is grounded in Toyota's own data: past engineering design reports, current regulations and handwritten documents by veteran engineers. This approach, retrieval-augmented generation, means the model looks up company documents before answering. Conversation histories and human experts' reviews of the AI's answers are stored so the agents can improve. Agents run in parallel, so adding more does not slow responses. Toyota's own blog post says the main challenge was getting each agent's answers accurate.
Results
Since January 2024 about 800 powertrain engineers have had access. The project lead says it is used hundreds of times a month. Toyota's blog post says it has not measured exact time saved, but users say searching for information became much faster. One engineer described the answer on exhaust-emission measurement equipment as detailed and accurate. Toyota plans more agents, such as one summarising common consumer complaints.
Lessons for your project
Target retiring expertise. Toyota's stated aim is to capture senior engineers' knowledge before they leave and pass it to the next generation.
Use many narrow agents rather than one general assistant. Each agent has its own data and tuning, and new agents can be added without slowing replies.
Ground answers in your own documents, including old design reports and handwritten notes from veterans, and have human experts review the AI's replies.
Decide early how you will measure value. Toyota had not yet measured time saved and relies on user interviews.
Caveats
- Both sources are Microsoft publications: a feature story and a blog post co-authored with a Toyota manager. No independent source confirms the figures.
- Toyota reports no time-saved or cost figure. The blog post says exact research time saved has not been measured; engineers say finding information is much easier.
- The usage figures are small and worded differently in the two sources (hundreds of times a month versus more than 100 requests a month). They are not reconciled here.
- The blog post describes an earlier build with four agents (battery, motor, regulations and system control). The feature story says nine agents. The blog post does not name a model version; GPT-4o comes from the feature story.
- The January 2024 date is when the feature story says engineers gained access. The Microsoft feature was read from an archived copy.
Sources
- Toyota is deploying AI agents to harness the collective wisdom of engineers and innovate faster
Microsoft Source Asia · 2024-11-19 · Vendor case study · news.microsoft.com
- How Toyota uses Azure Cosmos DB to power their multi-agent AI system for enhanced productivity
Microsoft Azure Cosmos DB Blog · 2024-11-19 · Vendor case study · devblogs.microsoft.com
Summarized in our own words from the sources above. Spotted an error? See corrections.