SHEET 02 · TRUST ARCHITECTURE
AI doesn't fail for lack of intelligence.
It fails for lack of a human layer.
Trust Architecture is the discipline of mapping that layer: who actually trusts the tools, who's performing trust, who's gone quiet and what to do about each of them.
FIELD RECORDING · GAIL WEINER
"I have spent over three years working with frontier AI systems every day. Not testing them in a lab, building with them, creating with them, mapping what works and what breaks. What I've learned is that the gap between AI capability and human comfort is a trust problem. And trust can be architected. That is what I do."
A-01
SITE SURVEY· THE TWO ARCHITECTURES
Every AI deployment has two architectures. One is drawn: the systems, the integrations, the licences. The other is never drawn anywhere: the humans who decide, every day and mostly in private, whether to route their work through the tools or around them.
The second architecture decides whether the first one was worth paying for. Implementation budgets go on integration and training. Almost nothing goes on the actual bottleneck, which is human willingness to hand over a decision.
That willingness isn't a mood or a training gap. It behaves in patterns - observable, repeatable, mappable. I've spent three years mapping them, inside organisations and inside my own practice, and the patterns hold from a five-person team to a five-thousand-person rollout.
A-02
THE CAST· YOU ALREADY KNOW THESE PEOPLE.
The taxonomy holds eleven behaviour types, each mapped to the engines running underneath it. Here are six you'll recognise from your own floor:
SUBJ-01
The Keeper
The one person who actually understands the tools, so everything AI-shaped routes through them. Looks like adoption on the dashboard. It's a single point of failure wearing a halo.
SUBJ-02
The Grammar User
Uses AI to tidy emails and polish documents. Nothing that touches judgement. Registers as an active user in every metric you have. The deployment hasn't reached them at all.
SUBJ-03
The Chaos Goblin
Experimenting constantly, loudly, everywhere. No workflow survives contact. Exhausting, and the closest thing you have to an R&D department, if you can make it visible instead of feral.
SUBJ-04
The Burned
Trusted a tool once and got hurt - an error with their name on it. Now everything gets re-checked by hand. The caution is rational. It's also permanent until someone works with the injury, not the behaviour.
SUBJ-05
The Anxious Self-Protector
Using the tools and hiding it, or avoiding them and hiding that. You won't find them by asking - self-report is exactly what they've learned to manage. Sometimes they're the only one reading the situation correctly.
SUBJ-06
The Calibrator
The end state. Trusts the tools for specific things, verified through their own testing, updated as the tools change. Most organisations make calibration unaffordable - then punish its absence on the dashboard.
That's six of eleven. The full taxonomy maps every type to the engines underneath, because the same behaviour can run on three different engines and each engine takes a different intervention. Treat the face instead of the engine, and the intervention backfires.
DETAIL A-02 · THE FACES ARE ON THIS PAGE. THE ENGINES ARE THE WORK.
A-03
METHOD · EVIDENCE, NOT ADJECTIVES
One rule runs through everything I do: I don't work from anyone's interpretation of anyone else. "She's resistant to AI" is an interpretation. "She writes her reports manually and re-checks anything the tool produces" is an observation. The work runs on the second kind only. That rule exists because of the two findings the whole discipline rests on:
FINDING 01 · THE CEILING
A leader's calibration is the deployment's ceiling
Teams don't adopt past the person running them. They watch what the leader actually does with the tools, not what the rollout plan says. So the work starts at the top, with the leader's own practice, before it goes anywhere near the floor.
FINDING 02 · THE IDENTITY
Self-report is identity performance
Ask people how they use AI and you get an account of who they'd like to be at work. So the work never runs on surveys or asking around. It runs on observable behaviour, collected deliberately, read against the taxonomy.
Put the two together and you get the method: map the leader first, then map the floor through structured evidence and read what the leader could and couldn't see as its own diagnostic. Two diagnostics, one room.
A-04
THE RULES I WORK BY
RULE 01
Nothing an individual tells me travels
Where the work extends to sessions with team members, what a person says in their session never goes back to their leader or anyone else. Leaders receive patterns and recommendations - never quotes, never verdicts on individuals. Written into my terms, stated to every participant before we start.
RULE 02
Types never travel to the floor
SUBJ-01
The taxonomy is a diagnostic instrument, not a vocabulary for labelling colleagues. No one on your team will ever be told what type they are - by me or by you.
RULE 03
No lab rats
SUBJ-01
Interventions are designed to read as management, not experiment. People who feel studied produce performance, not data - which corrupts the diagnosis and, more to the point, isn't a decent way to treat a team.
RULE 04
Complete in itself
SUBJ-01
Whatever engagement you buy stands whole. If it surfaces further work, I'll say so as a finding and you decide. Nothing is designed to make you need the next thing.
A-05
FROM THE FIELD
CLIENT RECORD
Gail got me comfortable with AI at home before we ever brought it near my work. Within months I'd gone from that to analysing what future job roles will look like in five years, and when I was asked to write the report on how our department should implement AI, her thinking helped me shape the approach, AI as a strategic partner to the humans, not just a screening tool. Gail doesn't teach you software. She changes how you and the AI work as a team.
- EXECUTIVE RECRUITER, TELECOM · 1:1 ENGAGEMENT
CLIENT RECORD
My company gave us a custom tool with every AI model in it, and honestly, I was afraid of the thing. Gail changed that. She taught me to just talk to the models instead of hunting for magic prompts, showed me how each one works differently, and got me using voice mode to think things through in a way that actually feels relaxing. I went from avoiding AI to it being the most natural part of my workday.
- MARKETING EXECUTIVE, PHARMACEUTICAL ADVERTISING · 1:1 ENGAGEMENT
A-06
ENTRY ·FOR SENIOR LEADERS
ENTRY · 1:1
The Trust Architecture Diagnostic
Three 90-minute sessions over roughly a month. I map your own calibration first, then your floor through evidence you collect, then hand you a prescription you can run without me. The company pays. The work is with you.
$2,000
£1,575 · DIRECT BOOKING · NO PROCUREMENT
A-07
SCALING THE WORK
FLAGSHIP · LEADERSHIP LINE
One method, across your leadership line.
The Diagnostic reads one leader and one floor. The Floor Map runs the same engine across three to five of your leaders, each gets the full Diagnostic, then I map the patterns across all of them into one synthesis: where the same engine is running on different floors, which interventions sequence across teams, and what your organisation's actual trust posture is, read from evidence rather than a dashboard. You get each leader's map, the synthesis, and one intervention sequence for the whole line.
Same rules, one addition the format needs: nothing an individual leader tells me travels to you or to each other. Patterns, never verdicts. Complete in itself.Most Floor Maps will start as one Diagnostic. Start there.
$12,000
£9,450
THREE TO FIVE LEADERS
DIRECTION BOOKING
NO PROCUREMENT
Start a conversation
I work with a small number of senior leaders and companies at a time. If you can feel the gap between what your adoption dashboard says and what's actually happening on your floor and you'd rather know what's in it:
Need engineering or QA capacity alongside the human-layer work? That's Sheet 04 - Engineering Capacity - curated European teams I've worked with for over a decade.