Question 01
“Your AI plan will automate thirty percent of the work. Thirty percent of what?”
Why no one can answer it: there is no denominator. An org chart is a map of authority. A job description is a map of intent. Neither is an inventory of the work itself — the actual, individual actions the organization performs. So every automation percentage anyone has ever presented is a number over an unknown. Including the ones in your own deck.
The only way anyone has answered it: build the inventory. NOW OS maps an organization's work across ten levels — from the institution down to the atomic action — so “thirty percent” finally has a whole to be thirty percent of.
See the ten-level work model →
Question 02
“Pick any task. Show me the evidence it belongs to a person, a machine, or both.”
Why no one can answer it: allocation today is decided by hierarchy, habit, budget, or whichever vendor demoed last. No task has ever been scored, so there is no evidence to show — only an opinion, delivered confidently. Ask it and listen to the room reach for one.
The only way anyone has answered it: score it. Every action is scored against your own rules — your policies, your risk appetite, your values — and lands in one of four modes: Automate, Augment, Assist, or Human-Only, with the reasoning attached. Not a binary, and never a verdict you can't inspect.
See the four modes →
Question 03
“Which decisions is AI making in your organization right now — and who signed for each one?”
Why no one can answer it: agents are already running. Nobody kept the list, and nobody recorded a named approver per decision. When a regulator, an auditor, or the board asks — and they will — “we'll pull it together” is the tell. It cannot be pulled together, because it was never written down.
The only way anyone has answered it: record it structurally, not after the fact. In NOW OS nothing promotes itself to autonomy; a named person approves each allocation, and every decision lands in a tamper-evident record that re-verifies itself every time it is read.
See the six structural guarantees →
Question 04
“What does one unit of that work cost by hand, versus by machine — and how do you know?”
Why no one can answer it: salaries and FTE counts exist. Cost at the level where work actually happens — the action — does not. Which means every AI ROI figure in every deck you've approved was a guess dressed as a number.
The only way anyone has answered it: compute it at the action. NOW OS carries a labor cost per action — done by hand versus done its way — and rolls it up. The ROI becomes a sum of things you can inspect, not an estimate you have to trust.
See a worked case with the numbers →
Question 05
“What work must stay human — and where is that list?”
Why no one can answer it: everyone has an automation list. Nobody has a defended human-only list with the reasoning attached — because “keep it human” has always been a feeling, not a finding.
The only way anyone has answered it: reserve it on evidence. Human-Only is one of the four modes, held to the same scoring as the other three. On the reference instance, tens of thousands of actions are recommended Human-Only — by the same rules that recommend automation elsewhere. We may be the only AI company whose product tells you what to keep human.
Why Human-Only is one of four →