The Hidden Infrastructure of Trust

I have a vivid memory of a meeting a few years ago. A team of clinicians was describing a workflow problem that had become a part of their daily work. There was nothing urgent that had caused the meeting. There had been no outage. There was no catastrophic failure. The application that they used simply did not support the way the work was done.

Gradually, a few steps began to separate. The documentation did not keep pace. A minor change upstream began to have a downstream effect that someone eventually noticed. What was most striking to me was the sense of normalcy that the situation created. No one had failed at anything. They had adapted, the way people in healthcare so often do when the solution is not quite a fit but the work still has to move forward.

The mismatches had quietly built up over time. This experience influenced the way that I understand the structures that surround a given tool. It was important in helping me understand the value of the trust that is lost when the systems that support a tool become disconnected. This is especially true as more AI moves into the work of healthcare, and even more so in a system that lacks that support.

What follows is about the structures that support the technology in healthcare, and the operational work that keeps it grounded in clinical reality.

What I Observe In Different Healthcare Contexts

The same challenges tend to be prevalent through large organizations, community and rural hospitals, and through various teams developing AI-integrated tools. Despite building tools that people can and want to trust, the surrounding systems and processes were not purposefully designed with technology that learns, updates, and changes over time.

Many challenges stem from informal and undefined ownership. When a model or policy is updated, many downstream teams are left with a lack of knowledge regarding the changes. It is true many clinicians provide observations and insights, but the feedback mechanisms often lack the necessary efficiency and clarity. The workflows and practices within a system will often change and evolve at a much quicker pace than the surrounding systems and structures. This is not a failure of the system. It is the inherent nature and complexity of the healthcare system. The introduction of AI simply highlights the existing gaps even more.

Where Trust is Formed

As a result of my leadership and advisory experiences, I have realized trust is formed in three specific, practical, and consistent areas. Regardless of the context, these areas are consistent and evident across the many diverse settings and systems I have been a part of, regardless of their specialty or size.

Structural Foundations

These elements are often invisible when functioning well. They include unambiguous ownership, a complete list of models and rules, and a standard methodology for examining alterations prior to deployment. In many of the organizations I have aided, the analytic teams produced exemplary work devoid of a common inventory system. Until the advent of AI, there was no compelling reason to have a common inventory system. AI is a game changer. It has increased the velocity and magnitude of changes for every alteration. In the absence of a common inventory, you have to trust a system that you cannot fully observe.

Day-to-Day Behaviors

Trust is also established during small instances of uncertainty. An instance is a clinician who observes something out of the ordinary and mentions it. Another is an analyst who notices a pattern that might be drift but is uncertain who should be notified. These instances are not system failures. Rather, they represent the absence of a clearly defined pathway for a signal to be communicated. When the structures around a tool mature, this is often the first place the change shows. Questions are answered more rapidly, and the signals are received by the appropriate individuals. The responsiveness of the system improves.

The Experience of Using It

Trust, or the lack thereof, is contingent upon the lived experience of the tool user. I have seen reasonably designed tools lose traction because of the friction or skepticism that the surrounding workflow created. One or two extra clicks, or an output that does not match what the clinician is observing, is friction, and even minor friction chips away at how reliable the tool feels. If the experience is seamless and friction is eliminated, the tool is more likely to be relied upon. Trust, unfortunately, is lost much more easily and silently.

The Quiet and Unobserved Drift

Many of the same patterns occur in the environments in which I work. A tool acts in an unexpected manner and someone adapts. The work goes on. No incidents are reported. No one stops to examine whether the unexpected behavior is merely a symptom of something more. These unexpected behaviors begin to accumulate over time.

This sort of drift is unremarkable. It does not trigger an incident. It slips past precisely because each instance is so minor. However, over time, the minor behaviors begin to diverge and create a greater separation between the tool and its intended design. The infrastructure of trust works to close that gap, allowing for minor deviations to be communicated to the developers before they become more significant. AI rarely, if ever, fails in a significant manner. It fails slowly, by letting small deviations go unobserved.

A Practical Indicator of Maturity

The most dependable indicator of how well the components of a tool are working together in an organization is the speed at which the organization can respond when something needs revisiting. In that context, speed is clarity. In that context, delay is confusion regarding ownership and prioritization.

The steps that make the biggest difference are not that complicated. Document ownership of each tool. Make it clear where a question is meant to be asked. Hold regular syncs of the clinical, operational, and analytical teams, so the people who see different parts of the picture actually talk. These are all small steps that, when implemented, make a substantial difference in reducing ambiguity. The value of the work is directly proportional to the reduction in time spent in ambiguity.

What Effective Oversight Has Looked Like in My Work

In differing contexts, meaningful progress is often the result of a few of the same practices. A reliable inventory of what is live, a clear front door for questions, and some structured collaboration to keep teams in sync.

These practices are quite simple, and that is the point. They enable an organization to be as fluid as the technology it is employing. They keep AI grounded in the reality of care delivery rather than allowing it to drift away, one small change at a time.

Importance of AI Integration in Healthcare

The integration of artificial intelligence (AI) in medicine is inescapable. Presently, there is a disparity between clinical practices and modern technologies. If necessary systems are not put in place, clinical work will unnecessarily be interrupted by technology, and things will fall apart at the most inopportune moments because of the widening gap between clinical practices and technology. The hidden infrastructure of trust narrows the gap between the technology and clinical practices. It builds trust and confidence in the new technology.

Without it, AI quickly becomes more of a disruption than a benefit, and that is the risk. AI work is not headline news and is done behind the scenes, yet everything builds from that foundation.

Christopher Hutchins Founder and CEO, Hutchins Data Strategy Consultants