An AI recommendation can arrive at a management meeting with two kinds of authority: the confidence of the answer and the position of the person presenting it. If the chief executive likes the proposal, an operational objection can be mistaken for resistance to change. A Norwegian company can then promise customers something its staff cannot reliably deliver.
The leadership decision is how to resolve a disagreement between an AI-assisted recommendation and the people responsible for carrying it out. My recommendation is to identify the disputed assumption and agree how to test it before asking the team to support the conclusion. Neither the tool nor the most senior person should settle a factual question by assertion.
A new survey raises a useful question
Charter's AI Pulse, released on 21 October 2026, surveyed 5,002 US desk workers using AI in August. Among 503 executive respondents, 32.8% said they trusted AI more than their own judgment, compared with 16.3% of individual contributors. Yet 81% of executives said they usually or always fact-checked outputs. The weighted, one-time survey measures reported attitudes, not decision quality or causation. It excludes nonusers and does not establish Norwegian prevalence. See report pages 32, 33 and 46.
The question for a Norwegian management team is therefore about its own practice: can someone challenge the chief executive's AI-assisted proposal without having to prove that AI is generally unreliable?
Checking references is valuable. It does not, by itself, establish that a recommendation fits the company's actual capacity. A market report can be accurate while the proposed delivery promise remains unrealistic.
The growth proposal and the duty roster
Consider a hypothetical Norwegian maintenance company, organised as an AS, assessing expansion into a neighbouring region. The chief executive uses an approved AI tool to explore demand, compare entry options and draft a proposal. The resulting recommendation favours a rapid launch with a short response-time commitment.
The service manager objects. Available technicians already cover existing contracts, and the proposed response time depends on travel conditions and backup staff that have not been confirmed. The commercial manager worries that delaying the launch will leave an opening for a competitor.
All three concerns deserve examination. Asking the AI to write a stronger growth case does not resolve the staffing question. Nor does the service manager's experience automatically establish that expansion is impossible. The disagreement concerns a testable condition: can the company meet the proposed promise while maintaining its existing service?
Management can ask the service team to check the proposed coverage against actual rosters, existing commitments and realistic travel times. It can ask the commercial team to establish which customers need the shortest response time and whether a narrower offer would still be attractive. Staff knowledge becomes part of the decision rather than an objection to be overcome after approval.
Use the company's approved information environment. There is no need to put named employees' schedules or confidential customer terms into an unrestricted chatbot to have this discussion.
Record the disagreement before resolving it
For a consequential decision, I suggest a short note that travels with the proposal. In the hypothetical company, it could read:
- Proposed commitment: Launch in the new region with the response time described in the offer.
- Disputed assumption: Existing staff and backup arrangements can cover both regions without weakening current commitments.
- Evidence needed: A coverage review based on the roster, travel conditions and existing contracts, plus confirmation of any external backup.
- Decision owner and deadline: The chief executive decides after the service manager and commercial manager have reported their findings, before an offer is sent.
- Outcome: Record whether the promise was retained, narrowed or postponed, and which evidence justified that choice. State what remains uncertain.
This is a proposed working method, not an intervention validated by the survey. Its purpose is to make a substantive objection visible and answerable. It should not become a requirement for routine emails or other low-consequence work.
Where the chief executive has already formed a strong view, ask those responsible for delivery for their assessment before circulating the preferred recommendation. They may use approved tools, but should identify their evidence and unresolved assumptions. Several polished memos repeating the same input are not several independent confirmations.
Keep the opportunity, change the commitment
The choice need not be a full launch or no expansion. The company could offer a longer response time, begin in a smaller area or secure a backup arrangement before taking on the most demanding customers. Each option changes the commercial proposition and needs to be checked with customers and the people delivering it. A constrained launch may create useful evidence sooner than an indefinite delay.
AI can help generate these alternatives and expose questions the team has missed. There is empirical reason to preserve that opportunity. A randomised study of 758 BCG consultants found gains in speed and quality on selected tasks, but worse correctness on a different managerial task. The experiment used GPT-4 in June 2023; the journal article appeared in March 2026. Its selected tasks and professional sample limit generalisation. It is not a benchmark of today's models or a test of Norwegian executive decisions.
My interpretation is that confidence in AI should be earned for the particular contribution it makes. Generating options, identifying missing information and recommending a binding customer promise are different contributions. Success with the first does not establish reliability with the last.
The chief executive owns the answer
Management should record how it resolved the material objection, including when it decided to proceed despite uncertainty. Do not ask the person who raised the issue to approve an unsupported assumption simply to demonstrate support for AI adoption.
For a small AS, this can fit into the ordinary management meeting and decision log. A larger company may need several functions to validate the proposal. The board's role, where the commitment is material, is to ask which assumptions management challenged, what changed and what exposure remains. Directors should not take over roster planning or run a parallel operational review.
The separate guides to AI in board preparation and developing AI skills at work address document handling and learning. This decision is about whose evidence reaches the meeting when a powerful recommendation is already on the table.
At the next significant AI-assisted proposal, ask: What is the strongest unresolved objection, who can establish whether it is right, and what would change our decision? A useful answer may strengthen the proposal. It may also make a smaller commitment the better business choice.
Sources checked on 23 October 2026. The maintenance company and decision note are hypothetical. The working recommendations are the author's analysis. The illustration is AI-generated.