Denso builds AI-powered platform to capture engineers’ tacit know-how

Denso’s project to mine and operationalize engineers’ tacit knowledge with AI is a glimpse of how large enterprises will actually use LLMs over the next few years. Instead of chasing flashy copilots, they are building RAG-centric platforms that ingest messy internal artifacts—meeting recordings, design rationales, failure analyses—and turn them into something searchable and composable. That’s exactly the kind of domain-specific context today’s general models need to deliver non-trivial productivity gains for high-skill work. ([dcross.impress.co.jp](https://dcross.impress.co.jp/docs/usecase/004448.html))

For the AGI race, these deployments matter because they quietly expose models to complex, real-world reasoning tasks that benchmarks don’t capture: reconciling conflicting expert opinions, navigating legacy design constraints, or explaining why a past decision was made under uncertainty. Each time an engineer uses the system and corrects or extends an answer, the organization is effectively fine-tuning a specialized “collective brain” on top of foundation models. At scale across hundreds of companies, that feedback loop could be as important as bigger base models or more GPUs.

Competitive dynamics are also interesting: vendors that can provide secure, on-prem or VPC-native stacks for this kind of knowledge management—vector stores, orchestration, governance—will gain deep embed in industrial supply chains, especially in Japan’s manufacturing base, which has historically been cautious about cloud and data sharing.