On this article
My work on AI starts with an organisational question: what can I responsibly delegate to an agent, and what evidence should I ask for before I delegate more?
That question connects three strands of work: NHAAO, a concept for a human-governed, agent-operated organisation; Olympus, where an assessment examined what could be carried forward from that earlier work; and the AI Operating Model, or AOM, through which I am developing a practical inquiry into governed autonomy. They belong to the same line of inquiry, but they do not have the same evidential status.
NHAAO expressed an ambition. The Olympus assessment supplied a design lesson. The AOM gives the ongoing work a way to distinguish authority, permitted effects and evidence of outcomes. None of those statements establishes a proven autonomous organisation or a demonstrated improvement in cost, speed or reliability.
Why the organisational question matters to me
My background is in business studies, and I completed MIT Sloan Executive Education's Implementing Agentic AI: Building Your Organizational Playbook. That context helps explain the questions I ask. It does not validate an answer, and the executive education course is not an MIT degree.
An agent can produce a convincing document while leaving the actual assignment incomplete. It can also complete a technically possible action without having authority to make that change. I want to examine how a delegation makes those distinctions explicit: who owns the decision, what may change, what happens when an exception arises, and what counts as completion.
These are questions about organising work as much as questions about models. A clearer contract, a simpler deterministic workflow or a better human handover may sometimes solve the problem more effectively than adding an agent. A useful research programme must leave room for that result.
What readers can expect
The AI Operating Model Field Notes will report bounded observations, changed assumptions and unresolved problems. A proposal will be labelled as a proposal. A test will identify what it observed and what it could not establish. Where a result is negative, that result should still be useful to someone trying to understand the same boundary.
Novasean provides a practical setting for this inquiry and publishes the notes. The scientific purpose remains separate from the hosting sales funnel. Buying a service does not validate an AOM claim, and engagement with an article does not demonstrate that governed autonomy works.
My commitment here is to make the reasoning open to challenge. When another approach is stronger, the useful response is to learn from it. When the evidence is too narrow to support a conclusion, the article should say so.
A small starting point
Take one task you might delegate and write down the expected outcome, the changes it may make and the evidence needed to accept the result. Then ask what would disprove your confidence in that delegation. Those questions do not create a tested system, but they make the next test more precise.
This introduction records the purpose of my work. It makes no change to the AOM and reports no new experiment. The first Field Note develops the move from an organisation-wide ambition towards bounded delegations; subsequent notes will examine the NHAAO and Olympus lesson and explain the operating-model questions in more detail.
Evidence note
The project history and qualifications above are author-supplied context recorded in the programme's foundation sources. They have not been independently credential-verified for this article. MIT's course description identifies the course; it is not evidence of my individual completion. The NHAAO and Olympus account is limited to the concept and the private, static design assessment described in the Field Notes, rather than an observed production organisation.
AI collaboration disclosure: AI assisted with drafting and source comparison. Maarten Strootman is the named author and accountable editor.