Public release candidate - mechanism proof only

The Azzam CEO operating loop

Azzam CEO is being designed as an owner-governed operating loop, not a free-roaming agent with unlimited authority.

The current implementation is deliberately narrow: a project-scoped Codex skill, persistent operating memory, driver-tree discipline, team and persona records, work items, decisions, artifact reviews, KPI observations, and confidence checks.

The loop

1. Load the business profile, driver tree, charter state, current focus, open work, recent decisions, team status, KPI observations, and relevant memory.
2. Confirm the operating mode: setup, informal advisory, or active CEO under an approved charter.
3. Work from the driver tree instead of treating every new idea as equal.
4. Dispatch explicit AI operating-team roles only when the owner or approved charter authorizes that work.
5. Produce an artifact, decision, work item, or reviewable recommendation.
6. Self-review the output for strengths, weaknesses, risks, assumptions, and recommended changes.
7. Record the evidence trail: what changed, what was reviewed, what was measured, and what still needs approval.
8. Use the next session to continue from that operating record instead of starting over.

The key design choice

The key design choice is owner governance. Azzam can help reason, draft, review, and coordinate approved AI team work, but public posting, spending, customer contact, destructive changes, credential handling, and sensitive claims still require owner approval.

We are not trying to make the AI sound more powerful than it is. We are trying to make the operating record inspectable enough that the owner can see what happened, what was assumed, what is still unproven, and what should happen next.