AIRedirect, don't resist
Work with the model's strengths and channel its weaknesses, so the interaction stays balanced and productive.
The story of AI-KIDO
A way of learning to work with AI, built patiently and tested on real projects. The repository is the toolbox that grew out of it.
A method first. A tool second.
AI-KIDO did not begin as a code repository. It began in January 2025 with a question: how do you learn to work with an AI assistant in a way that is reliable, repeatable and easy to pass on? The answers came one experiment at a time, on real projects. The repository arrived in May 2025 to hold what the method was producing, and the first numbered version followed in November. Today's toolbox stands on those months of learning.
21months of work, Jan 2025 → Oct 2026
5numbered versions, 1.2 to 1.4
15technology stacks covered from the first repository
MITopen license since the public release
Born from a simple observation
Faced with the raw power of language models, many developers swing between wonder and frustration. Like a beginner facing a stronger opponent in a martial art, they lose their balance: sometimes they dominate the exchange, sometimes they are carried away by the quirks of the AI. The name says it all. AI-KIDO joins "AI" with Aikido, the way of the harmony of energies. Instead of fighting the limits of an AI, you learn to use its energy and turn apparent weaknesses into strengths.
Let documentation lead
From test-driven to documentation-driven development: what is written down becomes the first source of truth and guides every exchange with the AI.
One source of truth
Truth tables stay flexible and evolve with the project, while validated decisions are locked as firm foundations. You decide what stays open and what is settled.
Always breathing
Regular cycles of review and improvement keep the method alive, and lessons learned flow back into it.
The real challenge is not learning to use AI, but learning to dance with it. The spirit of AI-KIDO
A path, not a manual
AI-KIDO is taught the way martial arts are: step by step, with belts that mark your progress. The guide that accompanies the method is built as a dojo. Each level rests on the previous one, so you advance at your own pace, from your very first prompt to the most sophisticated orchestrations.
White belt
Foundations
Set up your environment and learn the eight principles that everything else rests on.
- The eight AI-KIDO principles
- A working environment and your first functional knowledge base
- Structured prompts and expert personas
- Data validation, tests and first optimizations
Outcome: You generate reliable code in your own technology from a structured prompt.
Yellow belt
Expansion
Master your assistant and get more out of every prompt, in any field and any technology.
- The five specialized modes: Code, Architect, Debug, Orchestrator and Ask
- Optimized prompts, with around 40% fewer tokens
- An advanced Memory Bank and community patterns such as SPARC
- Plugging external tools in through MCP
Outcome: You spend fewer tokens and pick the right mode for each task.
Orange belt
Mastery
Move from individual practice to a base shared by the whole team, with rules that hold.
- A collaborative structure organized by roles, technologies and projects
- Team standards and governance, with hard points that lock what is decided
- Automatic synchronization of the shared base
- Metrics and a dashboard to follow adoption
Outcome: The whole team works from a single source of truth.
Green belt
Professional practice
Theory becomes everyday practice, from kicking off a project to passing the knowledge on.
- Katas to start a project, generate code, review, document and test it
- Applications for each role and fully integrated workflows: CI/CD, Git hooks, automated validation
- A persistent Memory Bank and multi-project orchestration
- Complex problem solving, innovation and mentoring
Outcome: You are autonomous on real projects and able to train others.
Blue belt
Agentic industrialization
Once your production chain is mastered and stable, the next step is industrialization with autonomous agents.
- Automated tests driven by specifications (SDD) and by tests (TDD)
- Autonomous AI-KIDO agents working along your production chain
- A new dimension for the developer, who guides the agents and dances with them
- A system that opens up to several teams and several projects
Outcome: Your production chain runs with agents, under your direction.
Black belt
Continuous improvement
The whole system keeps getting better, and you help it along.
- Fewer tokens spent and greater efficiency
- Strict respect of the chain, from end to end
- Constant innovation that keeps the method alive
- Contributions back to the method itself
Outcome: You no longer only use the method: you make it evolve.
The dojo spirit guides the whole journey: patience, perseverance, humility and curiosity. The goal is not "AI that codes for me". It is the augmented developer, who combines human expertise with the power of language models.
The same path also exists as a hands-on training in five workshops, from the first prompt to full autonomy: foundations, mastering the assistant, a knowledge base shared by the whole team, integrated workflows and automation, then expertise and mentoring. Each workshop has its own exercises, corrections and a review of what was learned, so every developer leaves able to apply the method and to teach it in turn.
The repository: the method made tangible
The AI-KIDO repository is not the starting point of the story. It is the result. Everything in it exists because the method needed it: a knowledge base organized by roles, use cases and techniques, specialized personas, ready-to-use workflows, and internal agents and commands that keep the whole thing healthy. Learning comes first, the toolbox follows, and both improve together.
Timeline
From the latest release back to the very first idea.
-
Kimono MCPThe Memory Bank opens up to connected clients
With Kimono, the AI-KIDO ecosystem gains a standalone MCP server: shared documents, commands, agents/personas, skills and workflows, indexed search with local semantic inference, and protected file and Git operations. One HTTP instance can serve several authorized clients around the same knowledge root.
Docker Compose supports standalone use; native stdio offers a dedicated connection. Platform integration, including Tatami, uses one instance per authorized workspace. Kimono supplies knowledge and instructions, not a built-in LLM; dynamic platform provisioning and routing remain a separate integration step.
Explore Kimono on GitHub · Capabilities and usage · Diagrams and setup tutorial · English usage guide
-
3 October 2026
Version 1.4The first public release
The whole Memory Bank now has an English baseline, the project ships under an MIT license and its configuration is cleaner and safer. The method and its toolbox are open to everyone.
-
27 February 2026
Version 1.3.3Skills with a full lifecycle
Skills can now be created, installed, updated and removed. The Jira and Confluence integrations leave the core and are published as skills in their own repository, so the method stays lean and each capability evolves on its own. A complete technical analysis plan joins the toolbox, and the method opens up to contributions from the team.
-
16 January 2026
Version 1.3.1Connected to team tools
Project tools such as Jira and Confluence are explored and built in, then connected through MCP servers, a cleaner and more durable approach than embedded scripts.
-
January 2026
The 1.3 architecture
A new organization gathers everything in one place. AI-KIDO becomes a real toolbox, with its own internal agents working together to maintain the Memory Bank.
-
1 December 2025
Version 1.2.1Ready for Claude Code
Compatibility with Claude Code is validated and consolidated, with a command that keeps the project's assistant configuration aligned with the Memory Bank.
-
21 November 2025
Version 1.2The first numbered version
A major restructuring for AI assistants: standard entry files for agents, creation and maintenance commands, a normalized base and the Light mode. The method is now ready to be used by others.
-
November 2025
The commands take shape
A persona taxonomy and a persona creation command, then project initialization, automatic maintenance of the Memory Bank and agent compilation. The rules of the method are consolidated, the Light mode returns for smaller teams, and the method starts speaking the language of today's AI tools.
-
October 2025
Refocus: one method, not a collection of copies
A deliberate strategic choice: stop keeping one copy of the method per project and concentrate everything in a single reusable Memory Bank. The project becomes lighter, clearer and ready to be shared.
-
August → September 2025
Training, publishing, observability
A training course on design patterns and anti-patterns brings the concepts and techniques layers to life, and is published as a complete 44-page website. An observability project follows, built on the same principles.
-
June 2025
A shared foundation, new terrains
The common base grows into a solid foundation reused by every project. The method is put to the test in a very different field: a fiction universe, to prove it works far beyond software.
-
Late May 2025
Teaching the tool to know itself
The documentation of the assistant's own environment is embedded in the Memory Bank, so the AI can consult it and use its tools correctly. A first base on anti-patterns and a shared common base follow.
-
25 May 2025
Beyond software
The method leaves the world of code with a first project for primary school teachers: supporting pupils, adapting to special needs, preparing parent meetings. The core directories of the Memory Bank, hard points and specifications, appear.
-
16 May 2025
First repositoryAI-KIDO gets a home
A first complete toolkit arrives: four ready-to-use workflows (code generation, code review, documentation, project initialization) declined for more than a dozen technology stacks: PHP, Node.js, Java, DevOps and application audit.
-
January → May 2025
OriginsThe idea takes shape
Before any tool, there was a way of learning: how to talk to an AI, how to give it a memory, how to make results repeatable. Truth tables, personas and workflows are designed and tried on real projects long before anything is published.
And now?
The story continues in the open. Try AI-KIDO on your own project, follow the releases and tell us what you build with it.