Skip to main content

Build systems people trust

Consistency beats novelty when the stakes are operational.

Illuminated arched bridge at dusk with Mount Fuji beyond
Your trail
Reading tools
Publication record
Author
Thierry Gilgen
Edition
1
Published
2026-07-20
AI assistance
Not recorded
Editorial review
Not recorded
Structured source record
No structured sources recorded. Inline citations are separate.
Edition change
Baseline recorded from existing publication
SHA-256
fdfe2cce833dbb2363af69bcc373cf4e6aab03ea30dae885c0539e13b88ac07c

A checksum identifies the recorded text and metadata. It does not certify the truth of a claim or the contents of external links.

Thierry Gilgen. Build systems people trust. Edition 1. 2026-07-20. https://www.thierry-gilgen-ict.ch/field-notes/build-systems-people-trust

Edition history

Thierry Gilgen. Build systems people trust. Edition 1. 2026-07-20. https://www.thierry-gilgen-ict.ch/field-notes/build-systems-people-trust

People do not trust systems because those systems are innovative.

They trust them because they behave predictably.

A payment is processed.

A document remains available.

A calculation can be explained.

A railway arrives.

A message reaches the correct person.

An operator can determine what happened when something goes wrong.

Trust accumulates through repeated, ordinary success.

This is uncomfortable for technology teams because novelty is more visible than consistency. A new feature can be demonstrated. Reliability is mostly noticed when it disappears.

But operational systems are judged less by their best moments than by their worst ones.

A clever interface does not compensate for lost data.

An advanced AI model does not compensate for untraceable decisions.

Automation does not create trust when nobody knows how to intervene.

Trustworthy systems share a few characteristics.

They communicate their limits.

They fail visibly rather than silently.

They allow important actions to be reviewed.

They maintain understandable records.

They provide humans with meaningful control.

They do not change fundamental behaviour without warning.

They are supported by people who accept responsibility for operating them.

This does not mean every system must be slow, conservative, or resistant to change.

It means change must be introduced without destroying the confidence on which the system depends.

Artificial intelligence makes this especially important. Probabilistic systems can be useful precisely because they handle ambiguity. But ambiguity inside the model must not become ambiguity about accountability.

Users should understand when they are interacting with AI.

Operators should know which version produced an output.

High-impact actions should have appropriate review.

Performance should be measured against real tasks, not only technical benchmarks.

Incidents should improve the system rather than disappear into private chat channels.

The strongest technology products eventually stop feeling innovative.

They feel dependable.

That is not a loss of ambition. It is the point at which invention becomes infrastructure.

A system earns trust when people can rely on it without becoming careless, understand it without becoming experts, and challenge it without losing access to the process.

Novelty may persuade someone to try a system.

Consistency gives them a reason to keep using it.

Read what comes next

Receive Thierry’s newsletter through Substack, or follow new Library volumes in your feed reader.

Leave a note in the margin

Your submission is stored privately until reviewed or deleted. Only an edited, accepted note can appear publicly. Do not include confidential information. Attribution is optional.