dP
AI Talks
Thought leadership · Designing for AI

Designing for intent, not commands

A working set of frameworks I use and teach for building trustworthy AI into enterprise software: how the interaction model changes, the concepts designers need to hold, the UI patterns that emerge, and the prompt scaffolding that gets a team from blank canvas to production.

Command-based
One path, one predictable result
Click
Navigate
Form
Result

If A, then B. Always. The interface can assume one shape because the output never varies.

Intent-based
One goal, a range of probable answers
“Show me policy spend”
Answer · 92%
Variant · 71%
Variant · 44%

The same input can return different output. The interface has to hold content of varying length, format and confidence — and let the user verify it.

01 · The core shift

Intent replaces command

Traditional UX is command based: click a button, navigate a menu, fill a form, and the output is highly predictable. AI introduces intent-based interaction, and every role in the system changes with it.

{{ s.k }}

{{ s.v }}

02 · Trust & recovery

Every failure needs a visible exit

A probabilistic system will be wrong in public. The design question is not whether it fails but whether the user can see the failure and act on it. I hold four failure modes and require a named recovery for each.

Fabrication
Confidently wrong

The answer reads clean but the underlying figure does not exist in the source data.

Recovery: every claim carries an inline citation to the row, table or document it came from — verification is one click, never a support ticket.

Low confidence
Probably right

The model returns a usable answer at 61% confidence with two plausible alternates behind it.

Recovery: show the confidence, show the alternates, and let the user promote one — ranked options instead of a single hidden guess.

Thin data
Cannot answer yet

The feature is on but the telemetry underneath is too shallow to model.

Recovery: state the requirement in plain numbers, then offer a real choice — a linear projection, or cancel. Never a confident fabrication.

Wrong intent
Answered a different question

The system resolved an ambiguous request to the wrong metric, period or entity.

Recovery: restate the interpretation as editable chips — metric, timeframe, filter — so correcting the read takes a tap, not a re-prompt.

03 · Concepts

Three mechanics every UXer must hold

You cannot design the surface of a system you do not understand underneath. Each concept carries a direct consequence for the interface.

  1. {{ c.num }}
    {{ c.t }}

    {{ c.b }}

    UX takeaway

    {{ c.u }}

04 · Patterns

Four ways AI meets the interface

Which vendors drove Q3 overage?
Three vendors account for 71% of it. Want the breakdown?
Ask a follow-up…
Conversational thread

A linear, text-based thread.

Best used for

Open-ended exploration, troubleshooting, complex querying.

Spend increased across the quarter. Procurement teams saw the largest delta versus plan and flagged it late.
✦ Rewrite with AI
Embedded in workflow

AI folded into an existing action — a “rewrite with AI” affordance surfaced over selected text.

Best used for

Enhancing a specific user action without breaking focus.

Assistant
Add a benchmark line?
Docs: cohort logic
Persistent companion panel

A side panel that watches the workspace and offers real-time suggestions or documentation.

Best used for

Complex enterprise tools, design software, IDEs.

Weekly spend Anomaly ✦
Detected without a prompt: spend up 31% week over week.
Ambient insight

Background processing that surfaces insights or anomalies automatically, with no prompt at all.

Best used for

Dashboards, analytics platforms, monitoring tools.

05 · Staff AI UX Lab

The 4 UI states of predictive forecasting

A framework I authored for making a probabilistic feature feel trustworthy inside an enterprise dashboard — from where the trigger lives to how the system degrades when the data is too thin.

CURRENT DATE {{ deltaLabel }}
{{ stepLabel }}
State 04: Human-in-the-loop panel
Model variables & scenarios

Adjust model variables without engineering a new prompt or breaking workflow.

Projection horizon {{ horizonLabel }}
Confidence interval strictness
“What-if” macro simulations
{{ s.n }}
{{ s.t }}

{{ s.b }}

Graceful degradation

Flip forecasting on for a metric with three weeks of history and the system must say so plainly: AI forecasting requires a minimum of 6 months of historical data to model seasonality. This view contains 22 days of telemetry. Then it offers a real choice — project a linear trend anyway, or cancel — rather than quietly producing a confident, wrong answer.

06 · Prompt blueprint

“Spend over time”

A reusable prompt pattern for deriving a distribution or comparison visualization. The structure is the point: data, objective, visual preferences, clutter reduction, output format — so the model returns something a designer can actually ship.

Prompt pattern
{{ promptText }}
07 · Measurement

What “good” means when output varies

Task success and time on task break down the moment two users get two different answers. These are the four signals I instrument instead, and the design decision each one drives.

Acceptance rate
68%

How often the output ships untouched. A low rate is a prompt-scaffolding problem, not a UI problem.

Edit distance
2.4

Average corrections before the user is satisfied. Rising distance means the interface is hiding the variables that matter.

Time to verify
9s

How long it takes to confirm the answer is true. Citations and drill-through are judged on this number alone.

Escalation rate
7%

Share of sessions that fall back to a human or a manual export. The honest measure of whether the feature earned its place.

08 · Agentic workflows

From one answer to many actions

When the system starts acting rather than answering, the design problem shifts from presentation to permission. Autonomy is earned step by step, and every irreversible action needs a gate the user controls.

Interpret request
Draft plan
Approval gate ✋
Execute
Audit trail
Show the plan first

The agent states what it intends to do, in what order, before it touches anything.

Gate the irreversible

Reads run free. Anything that writes, sends, spends or deletes stops for a human.

Leave a trail

Every step is logged, attributable and reversible — the difference between a tool and a liability.