From a book about six hats to an engine for decisions
AITS is a theory that found its way into a repository, and each version was born from something the previous one could not do.
1985
Six Thinking Hats
Edward de Bono proposes thinking from one perspective at a time, six hats to wear in sequence and integrate at the end.
June 2025
The first articles
The blog publishes AI-Enhanced Thinking Hats (28 June) and the next day the first description of the Adaptive Intelligence Thinking System, with eight cognitive agents. (Italian)
October 2025
The book
La mente adattiva. Pensare insieme alle macchine (The Adaptive Mind, Egea) is published: the hats become agents that can be people, models or pairs of both, with a ten-point ethical manifesto. (Italian)
February 2026
Version 1
The book becomes open source code inside Claude Code, with the agents as markdown sub-agents, the first three commands, the playbooks and an HTML dashboard.
17 April 2026
Version 2.0
Agents grow from eight to eleven with Systemic, Foresight and Memory, plus JSON schemas for every output, the conflict matrix, shared taxonomies, the pattern library, three human oversight modes and seven inviolable rules.
6 November 2026
Version 3.0.0
Rules leave the prompt and move into a deterministic TypeScript engine, the spec stops naming models, and the release adds a hash-chained log, diversity across model families, semantic memory, calibration and four adapters.
7 November 2026
Version 3.0.1
The fixes found in real sessions: the log of record is written by the runtime through the SubagentStop hook rather than by the orchestrator, commands become /aits:full and so on, and gates work in sessions with nobody in front of the screen.
3.1 roadmap
The Brescia simulation showed that each new agent reporting the same data gaps reopened a gate already closed, and one quick-mode decision piled up twelve of them. A system that asks for confirmation too often teaches people to confirm without reading.
- a human acceptance that covers the same gap in later outputs too
- a gate answer that says "change plan" instead of "proceed"
- a clear choice on how many times critical agents run in quick mode
- real cases in the eval/ harness, to measure with numbers how well calibrated the system is
If you have a decision already taken whose outcome you now know and want to see what AITS would have said about it, get in touch through the contact page.
Changelog (GitHub)
This section reads CHANGELOG.md from the repository, so it always matches the latest release.
Loading… Read CHANGELOG.md on GitHub