One-Soup One-Dish: A Japanese Cooking Principle for AI Fatigue

Developer and creator Takuya, writing on his blog Devas Life, draws a parallel between the Japanese cooking principle Ichiju Issai (one soup, one dish) and the overwhelming pace of AI tooling. The core idea: strip away the unnecessary to maintain mental health and creative focus.
The Problem: AI Fatigue
New AI services emerge daily, and major releases from big companies change workflows weekly. Chasing every hype doesn't increase security — it clouds the essential skills developers need. Takuya calls this AI fatigue.
The Solution: Ichiju Issai for Tech
Ichiju Issai is a meal style centered around rice, one soup, and one side dish. Chef Yoshiharu Doi proposed it to free home cooks from the pressure of creating elaborate meals. Takuya applies this to software: treat your core development environment (editor, AI coding agent, or framework) as the rice — the constant that never changes — and allow yourself only one primary tool and one secondary tool. The goal is to create a rhythm where you can return to a comfortable, heart-centered place.
Practical Takeaways
- Decide what NOT to do. Doi says: “By simplifying meals to Ichiju Issai, cooking becomes free of stress.” Your tech stack should follow the same rule.
- Find tools you never get bored with. Just as a parent’s cooking brings comfort through familiarity, your core tools should feel like home.
- Avoid tracking gossip or drama. Social media algorithms try to grab attention; Takuya advises deliberately ignoring them to preserve calm.
- Direction over strategy. Moats can be flexibly changed; the important thing is knowing where you want to go as a developer and artist.
Takuya’s article includes a video version and references Doi’s book 一汁一菜でよいという提案 (The Proposal for One Soup, One Dish).
📖 Read the full source: HN AI Agents
👀 See Also

Practical Lessons from Building On-Device AI in React Native
A developer shares specific technical details from building a React Native app with on-device LLMs, image generation, voice transcription, and vision AI, including memory management strategies, library choices, and performance benchmarks.

Getting the Most Out of Claude: A Data Analyst's Workflow with Cowork and Claude Code
A data analyst with no coding background shares how they use Cowork for end-to-end automation and Claude Code for heavy lifting — building a lead gen tool using Google Places API, a fraud dashboard, and automated social media posting.

Opus 4.7 Broke 40% of Prompts; Fix Was Structuring CLAUDE.md and Skills
After Opus 4.7 degraded ~40% of prompts across 6 setups, a fractional head of AI fixed it by replacing ad-hoc prompts with structured Skill files, hierarchical CLAUDE.md, and separate memory files — reducing token usage 22% and iteration turns from 3-4 to 1-2.

Java Performance Optimization: Eight Anti-Patterns That Slow Down Your Code
A Java order-processing app improved from 1,198ms to 239ms elapsed time, 85,000 to 419,000 orders per second, and 1GB to 139MB heap usage by fixing eight common anti-patterns identified through Java Flight Recording profiling.