Stuck in Pilot Mode: What Aviation Knows About Trusting Machines
McKinsey’s HR Monitor 2026 contains a sentence that should stop every people leader mid-page: many organizations appear stuck in pilot mode.
The integration of AI into human resources is moving sluggishly due to missing frameworks of trust, rather than technological constraints. Only 28% of HR processes currently have operational AI, while 37% remain stuck in testing phases. In Europe, operational adoption is just 23%, with 39% of processes lacking any AI solutions.
What makes this interesting is not the pace. It is the reason for the pace.
Asked how much of their function could be automated, HR professionals have reduced their own estimates from 30% down to 20% this year. The primary hurdles are governance ambiguity, disjointed data architectures, and capability deficits. Insufficient technology does not appear on the list.
The constraint is not what the machine can do. It is that we have not built the structure that would allow us to trust it at scale.
Aviation has met this problem before.
Our industry did not integrate automation into safety-critical work by being braver than everyone else, and it did not do it quickly. It did it by building an architecture of trust and then refusing to deploy anything outside it:
Certification before service entry: No system reaches an aircraft because it performed well in a demonstration. It reaches an aircraft because it was tested against a defined standard.
Licence scope: A type rating is not a general permission to fly; it authorizes a specific type, in specific conditions, with specific limitations.
Supervised operation: An agent validated on curated data has completed simulator training. It has not flown the line. Line training sits in between because performance under controlled conditions does not predict performance under real ones.
Recurrent checks: Aviation does not assume that competence, once demonstrated, persists. Systems that learn and adapt present a sharp risk: a slow drift in behavior that nobody intended and nobody noticed.
Command authority: An aircraft may be flown almost entirely by its systems, but it still has a commander. Responsibility does not dilute in proportion to how much of the workload was automated.
Where the Analogy Stops Being Comfortable
Aviation also learned, expensively, that automation generates its own failure modes: Crews trusting a system beyond its envelope. Losing track of which mode it is in. Losing the manual skill they were supposed to be supervising it with.
Every one of those has a people-function equivalent. A manager who accepts a recommendation because it arrived with a high confidence score. A team that quietly loses the ability to make the judgment the agent was meant to support.
The response in aviation was never to remove the automation. It was to train deliberately for the moment it hands control back.
What This Means for the People Profession
When an AI agent contributes to a decision affecting someone's role, pay, performance record, or employment, there must be one identifiable human being who owns that decision and can be asked to justify it. Not a committee, not a process, not a vendor.
Aviation's largest safety gains over the past fifty years did not come from better machines. They came from studying the interface between the human and the machine. Human factors is not a supporting discipline in aviation safety. It is the discipline.
The question in front of us was never whether an AI agent can be hired. It is whether we can define its scope, evidence its competence, detect it drifting, and preserve one unambiguous human owner of the outcome.
We do not leave pilot mode by being braver. We leave it by being certified.
You may also read the full article on LinkedIn https://www.linkedin.com/pulse/stuck-pilot-mode-what-aviation-knows-trusting-machines-yi%C4%9Fid-acar-qchlf
