A field manual for the AI era
Software only did what users made it do, so designers won by making products easy to operate. Now AI does the work — and users decide in seconds whether they trust it. Designers win by making that work easy to judge.
Design trust into every surface
Keep it simple — motivation is the product
Write decisions humans and agents both build from
Use this site if you are
The balance flipped
AI collapsed the gulf of execution. The effort didn't disappear — it moved. Every unit of work the AI absorbs comes back to the user as a harder question: did it do this well, and how would I know?
2
gulfs — execution and evaluation
1
short window to prove yourself
4
trust patterns to design
7
interaction patterns worth stealing
10
field guides in this series
The person
There's no roadmap to execute in a founding role. The designer helps write it — and builds the product while writing.
CAP / 01
Streaming states, recovery flows, confidence displays — these can't be judged as static frames. Designers who iterate on live components with tools like Claude Code run ten loops while mockup-bound designers schedule a review.
In practice: a working prototype of one interaction, this week, beats a deck about it.
Decide what to design next — and say why
Write
Prototype vertically
Streaming states, recovery flows, confidence displays — these can't be judged as static frames. Designers who iterate on live components with tools like Claude Code run ten loops while mockup-bound designers schedule a review.
In practice: a working prototype of one interaction, this week, beats a deck about it.
Evaluation design
Version A drafts a client email and shows Send. Version B drafts the email, links the two facts it pulled from the CRM, flags “not sure about the renewal date — worth checking,” and offers Preview → Send with one-step recall. That's evaluation design, and it's the difference between a second session and a quiet uninstall.
Read the gulfs guideVersion A
Version B
Version B answers the user's actual question — not “can it write?” but “can I stake my name on this?”
In order
Shape the product around the intent of the audience, guide users through patterns that build trust and motivation, and make progress feel fast.
By intersecting a real struggle with a promise specific enough to feel personal.
By converting it into motivation — early, felt progress on the problem the user came with.
Progressive disclosure — show the one thing that solves today's problem and let mastery unlock the rest.
From the field guide
01 / 04
The Bottleneck Moved
Field guide
01
“A working prototype of one interaction, this week, beats a deck about it.”
The comparison, plainly
The foundations did not change. Jobs-to-be-done thinking and UX discipline are the reason any of the new techniques work. What changed is where the work matters most.
the user's actual question
did it do this well — and how would I know?
Foundation
Deliverables
Communication
Trust
Motivation
Optimized the gulf of execution
Builds from the gulf of evaluation
Designed screens and flows
Designs behavioral flows, trust surfaces, and instructions
Delivered Figma files; engineers translated
Delivers working slices; engineers and agents extend
Specs and meetings
Prose that machines also execute
Earned trust slowly through reliability
Designs trust explicitly, surface by surface, from session one
Fought for usability
Fights for motivation — demotivation is churn
The series
The supporting series unpacks each pillar of the argument.
The core argument. Execution collapsed, evaluation became the bill, and verification design is now the moat.
The four trust patterns (evidence, honest confidence, reversibility, graceful recovery), progressive trust, and a 20-minute trust audit.
Figma for intent, Claude Code for behavior, markdown instructions for judgment — with a real example of what an instruction file looks like.
Why static state matrices fail, how vertical slices with realistic dummy data become the spec, and what this changes about handoff.
Getting attention, keeping it through felt progress, progressive disclosure, and the one-screen audit that predicts retention.
A decision guide for both sides of the table: scorecard, green/red flags, interview kit, self-assessment, and a first-30-days plan.
Seven interaction patterns worth stealing (receipts, ghost text, preview-then-commit, earned autonomy) and the anti-patterns to name and kill.
Iteration when building is nearly free: the five-day loop, the kill rate, and why shipped slop spends user trust.
When everyone rents the same models, judgment is what's left to own — and how to train it and write it down.
Users punish silence, not latency. Streaming, named work, partial results, and honest progress — the wait is a design surface.
Trust surfaces
A probabilistic system can't earn trust passively. Evidence, honest confidence, previews, undo, graceful recovery — trust is a set of interface decisions with specific locations on specific screens.
Cite sources inline, link evidence to the claims the output makes.
Say 'not sure about the renewal date — worth checking' instead of confident nonsense.
Make actions previewable and reversible; consequences come after consent.
When the system is wrong, show the recovery — that's where trust compounds.
Answers / index
Still deciding?
Both sides of the hiring table are working from outdated specs. The decision guide covers scorecard, flags, and the first 30 days.
Read the hiring guideThe real test
Quickly, visibly, and on every surface the user touches.