DIOS: a decision intelligence system for C-suite management
Michael S.U. Hudson designed and directed DIOS, a private decision intelligence system for C-suite management. It gives leadership a searchable memory, a tracked record of every decision and commitment, and a governed AI layer that supports decisions without ever making them.
Outcome
5 of 5
Past decisions retrieved with their context, in under a second.
Baseline
0 of 3 retrieved within two minutes, 15 June 2026
Result
5 of 5 at the gate test, 17 June 2026, under 0.3 seconds per query
Source
DIOS phase gate records
Situation
In most leadership teams, decisions, promises and context live in a few people’s heads, spread across meetings, chat, email and documents. Discovery at Project 54 measured it: three past decisions requested on demand, none found within two minutes. Commitments made in meetings had no tracker.
Role
Michael conceived DIOS, defined its purpose and constraints, and wrote the governance it was built under. He made every architecture and scope decision and approved every phase gate. The build was executed by AI agents inside that framework: they proposed, he decided.
What was built
Rules before code: a written constitution, three tiers of decision rights (AI acts; AI proposes and leadership approves; leadership only), a single source of truth for each of 13 data domains, a duty for the AI to challenge any over engineered or assumption driven recommendation, and kill criteria written in from the first day.
The system: meetings, chat, email, documents and notes are absorbed into a self hosted database and vector memory on owned servers, with no public access. Decisions are logged with options, rationale, what was unknown, and reviews at 30 and 90 days. Commitments are extracted from meetings and closed only when a later transcript proves completion. A segmented briefing arrives each morning.
Reliability: an eight rule write standard that stops AI errors reaching the system of record, including verbatim checking of every quoted piece of evidence, audit trails on every automated write, and validation against a labelled test set before writes are enabled. A 32 check self test and monitoring that catches jobs that succeed while capturing nothing.
Outcome
Seven days from programme start to a verified working system, for an estimated 8 to 12 hours of leadership time. About 30,000 items of institutional memory. Modelled running cost under £40 a month. When usage fell while client work took priority, DIOS reported it in its own weekly review, and on 25 August 2026 it was paused on the grounds of non use, with all data kept and a resume time of under an hour.
What it shows
AI can carry the build when governance carries the judgement. A system designed with the conditions for its own shutdown, and honest enough to report that it was not earning its keep, is one leadership can trust.