# Japanese companies are lagging behind on AI because they > Source: > Published: 2026-08-13 08:44:50+00:00 # Japanese companies are lagging behind on AI because they ## The cultural friction in AI workflows The primary bottleneck isn't the technology—it's the legacy mindset. Most Japanese firms operate on a consensus-based decision-making process (ringi), which is the absolute opposite of what prompt engineering requires. AI thrives on rapid experimentation, failing fast, and tweaking prompts in real-time. When you have to get five different department heads to sign off on a prompt change, you lose the agility needed to actually optimize an AI workflow. Furthermore, there is a deep-seated reliance on rigid, manual documentation. Many firms are trying to force AI to fit into their existing, archaic paperwork systems rather than letting the AI redefine how the work gets done. Instead of building a beginner-friendly internal knowledge base that an LLM can query, they're trying to digitize PDFs from 1998 and wondering why the [RAG](/en/tags/rag/) (Retrieval-Augmented Generation) performance is terrible. ## The talent gap and the "Expert" trap There's also a weird paradox where Japanese firms want "AI experts" to lead the charge, but they don't want to empower those experts to actually change things. I've seen cases where a data scientist provides a practical tutorial on how to automate a reporting process, only for the management to reject it because it doesn't follow the "traditional" way of auditing. To actually move the needle, these companies need to shift toward a more decentralized deployment strategy. Here is a basic framework for how a legacy firm should actually approach a deep dive into AI integration: 1. **Identify a "Low-Stakes" Win:** Stop trying to automate the core product first. Find a back-office task (like email sorting or internal FAQ) where a mistake won't crash the company. 2. **Build a Sandbox:** Create a segregated environment where employees can experiment with prompts without needing managerial approval for every single query. 3. **Iterative Feedback Loops:** Move from a "Waterfall" deployment to an agile one. Deploy a version that is 70% correct, then use human-in-the-loop feedback to get it to 90%. If they keep waiting for a "perfect" solution, they'll find themselves in a position where their competitors aren't just faster, but are operating with a completely different cost structure. The real risk isn't a hallucinating chatbot; it's the systemic obsolescence of a workforce that doesn't know how to collaborate with an agent. [Next Sign language AI finally works on a mobile device →](/en/news/6150/)