How I use AI in my everyday work
How I use AI in my everyday work
I’ve built a small operating system around AI: specialized agents, reusable design knowledge, structured review loops, and shared context that turns repetitive design work into leverage.
I’ve built a small operating system around AI: specialized agents, reusable design knowledge, structured review loops, and shared context that turns repetitive design work into leverage.
It’s faster exploration, sharper reviews, more consistent execution, and more time for the decisions that actually need product judgment.
It’s faster exploration, sharper reviews, more consistent execution, and more time for the decisions that actually need product judgment.
01 - PHILOSOPHY
01 - PHILOSOPHY
AI handles the repetition.
I keep the judgment.
AI handles the repetition.
I keep the judgment.
My goal is not to outsource product thinking. It is to remove low-value friction around it. I use AI to create options, pressure-test decisions, maintain context, review details, and make reusable knowledge available exactly when it is needed.
My goal is not to outsource product thinking. It is to remove low-value friction around it. I use AI to create options, pressure-test decisions, maintain context, review details, and make reusable knowledge available exactly when it is needed.
What I deligate
What I deligate
First drafts, copy variants, heuristic checks, edge-case prompts, consistency scans, documentation, summaries and repetitive review work.
First drafts, copy variants, heuristic checks, edge-case prompts, consistency scans, documentation, summaries and repetitive review work.
What I don't deligate
What I don't deligate
Problem framing, prioritization, user empathy, trade-offs, product taste, interaction judgment, and the final call on what is good enough to ship.
Problem framing, prioritization, user empathy, trade-offs, product taste, interaction judgment, and the final call on what is good enough to ship.
03 - MY AI LOOP
A repeatable system,
not random prompting.
A repeatable system,
not random prompting.
Instead of starting every conversation from zero, I give AI clear roles, persistent context, expected outputs and review criteria. That turns one-off prompting into a design workflow.
Instead of starting every conversation from zero, I give AI clear roles, persistent context, expected outputs and review criteria. That turns one-off prompting into a design workflow.
UX Copy Agent
UX Copy Agent
Generates and critiques interface copy using product context, tone-of-voice rules and UI constraints without treating copy as decoration.
Generates and critiques interface copy using product context, tone-of-voice rules and UI constraints without treating copy as decoration.
Writes concise labels, helper text, errors and empty states
Writes concise labels, helper text, errors and empty states
Creates variants for different intent and space constraints
Creates variants for different intent and space constraints
Checks clarity, tone, consistency and actionability
Checks clarity, tone, consistency and actionability
Helps unblock copy iteration without creating a dependency
Helps unblock copy iteration without creating a dependency
Design Review Agent
Design Review Agent
Acts like a structured critique partner. I use it after I have a direction
not before to expose blind spots and force clearer reasoning.
Acts like a structured critique partner. I use it after I have a direction
not before to expose blind spots and force clearer reasoning.
Reviews hierarchy, flows, states and interaction logic
Reviews hierarchy, flows, states and interaction logic
Flags weak assumptions and missing edge cases
Flags weak assumptions and missing edge cases
Challenges whether the solution solves the actual problem
Challenges whether the solution solves the actual problem
Separates high-impact issues from cosmetic feedback
Separates high-impact issues from cosmetic feedback
Consistency Guard
Consistency Guard
A final-pass reviewer that checks whether the design still behaves like the same product after dozens of screens, states and iterations.
A final-pass reviewer that checks whether the design still behaves like the same product after dozens of screens, states and iterations.
Checks component and pattern consistency
Checks component and pattern consistency
Flags terminology drift and mismatched interaction behavior
Flags terminology drift and mismatched interaction behavior
Looks for missing states and one-off design decisions
Looks for missing states and one-off design decisions
Acts as a pre-handoff quality gate
Acts as a pre-handoff quality gate
Edge-case & Failure Agent
Edge-case & Failure Agent
I use this agent to deliberately attack the happy path and ask: what breaks, what is unclear, and what will users do when the system does not behave perfectly?
I use this agent to deliberately attack the happy path and ask: what breaks, what is unclear, and what will users do when the system does not behave perfectly?
Generates realistic failure and recovery scenarios
Generates realistic failure and recovery scenarios
Surfaces permissions, loading, latency and empty-data cases
Surfaces permissions, loading, latency and empty-data cases
Tests destructive and irreversible actions
Tests destructive and irreversible actions
Improves the completeness of flows before engg handoff
Improves the completeness of flows before engg handoff
04 - RESUABLE INTELLIGENCE
The useful part isn’t the prompt.
It’s the context behind it.
The useful part isn’t the prompt.
It’s the context behind it.
I maintain reusable Markdown files that encode product knowledge, design rules and expected behaviors. They give AI the same baseline repeatedly instead of making every session depend on what I remember to explain.
I maintain reusable Markdown files that encode product knowledge, design rules and expected behaviors. They give AI the same baseline repeatedly instead of making every session depend on what I remember to explain.
design-system.md
design-system.md
UI rules
UI rules
ux-copy.md
ux-copy.md
voice + patterns
voice + patterns
product-context.md
product-context.md
domain knowledge
domain knowledge
review-checklist.md
review-checklist.md
quality bar
quality bar
agent-instructions.md
agent-instructions.md
specialist roles
specialist roles
Markdown files become a lightweight
design-memory layer
Markdown files become a lightweight
design-memory layer
These files capture the things that usually live in a designer’s head: naming
conventions, interaction rules, component expectations, product vocabulary, copy
principles and review standards.
These files capture the things that usually live in a designer’s head: naming
conventions, interaction rules, component expectations, product vocabulary, copy
principles and review standards.
Because the knowledge is reusable, the benefit extends beyond a single prompt or
a single designer. The same context can support faster reviews, clearer copy, more
consistent outputs and less repeated explanation across the team.
Because the knowledge is reusable, the benefit extends beyond a single prompt or
a single designer. The same context can support faster reviews, clearer copy, more
consistent outputs and less repeated explanation across the team.
05 - EXAMPLE IN PRACTICE
One flow. Multiple layers of AI assistance.
One flow. Multiple layers of AI assistance.
I don’t ask one model to solve the whole problem. I use different AI passes at different moments, with my own product decisions sitting in the middle.
I don’t ask one model to solve the whole problem. I use different AI passes at different moments, with my own product decisions sitting in the middle.
BEFORE DESIGNING
BEFORE DESIGNING
Frame and expand
Frame and expand
I use AI to expose assumptions, enumerate
states and explore alternate ways the
problem could be understood.
I use AI to expose assumptions, enumerate
states and explore alternate ways the
problem could be understood.
WHILE DESIGNING
WHILE DESIGNING
Generate and challenge
Generate and challenge
Specialist agents help with copy, edge cases
and critique while I make the interaction and
visual decisions.
Specialist agents help with copy, edge cases
and critique while I make the interaction and
visual decisions.
BEFORE HANDOFF
BEFORE HANDOFF
Audit and clarify
Audit and clarify
The consistency guard and handoff pass
check the small details that become
expensive when discovered during
implementation.
The consistency guard and handoff pass
check the small details that become
expensive when discovered during
implementation.
06 - WHAT CHANGED
The real output is leverage.
The real output is leverage.
The system matters because it changes how I spend design time. Repetitive work gets compressed; critique happens earlier; reusable knowledge reduces restart cost; and more of my attention can go into judgment.
The system matters because it changes how I spend design time. Repetitive work gets compressed; critique happens earlier; reusable knowledge reduces restart cost; and more of my attention can go into judgment.
Faster iteration
Faster iteration
More viable directions, copy options and
review feedback without adding more
coordination overhead.
More viable directions, copy options and
review feedback without adding more
coordination overhead.
Stronger consistency
Stronger consistency
Shared rules and automated checks make it
harder for product language and interaction
patterns to drift.
Shared rules and automated checks make it
harder for product language and interaction
patterns to drift.
Team-scale reuse
Team-scale reuse
Reusable context turns individual design
knowledge into something others can access
and apply.
Reusable context turns individual design
knowledge into something others can access
and apply.
06 - WHAT I'M CAREFUL ABOUT
Speed is only useful if the output is still trustworthy
Speed is only useful if the output is still trustworthy
AI can produce confident nonsense, flatten nuance, or make an average solution sound polished. So I treat every output as material to inspect, not authority.
I verify against the product
I verify against the product
AI suggestions are checked against the actual flow, existing patterns,
constraints and user intent.
AI suggestions are checked against the actual flow, existing patterns,
constraints and user intent.
I keep roles narrow
I keep roles narrow
Focused agents are easier to evaluate than one giant prompt pretending
to be researcher, writer, designer and PM at once.
Focused agents are easier to evaluate than one giant prompt pretending
to be researcher, writer, designer and PM at once.
I separate evidence from interpretation
I separate evidence from interpretation
Especially in research and feedback synthesis, I keep source material
available so a neat summary never replaces the actual evidence.
Especially in research and feedback synthesis, I keep source material
available so a neat summary never replaces the actual evidence.
I own the final decision
I own the final decision
The quality bar, taste and trade-offs remain mine. AI increases
throughput; it does not inherit accountability.
The quality bar, taste and trade-offs remain mine. AI increases
throughput; it does not inherit accountability.
THE POINT
“I’m not trying to become a
designer who uses AI.
I’m building a design practice
that is AI-native.
“I’m not trying to become a
designer who uses AI.
I’m building a design practice
that is AI-native.
For me, that means turning good judgment into repeatable systems: specialist
agents, reusable knowledge, faster feedback loops and stronger quality checks.
The technology changes quickly. The advantage is knowing where it should
accelerate the work and where a designer still needs to think.
For me, that means turning good judgment into repeatable systems: specialist
agents, reusable knowledge, faster feedback loops and stronger quality checks.
The technology changes quickly. The advantage is knowing where it should
accelerate the work and where a designer still needs to think.
