Expertise and AI integration
A field guide for design teams
AI + Design: how we work with intelligence
How AI and design come together in my work: a mindset for collaborating with AI, a way of designing that collaboration around context, and the systems that make it repeatable.
That shapes how design teams work with AI day to day, and how the products we design use context to behave well. It comes from leading AI design at Comcast and an MIT xPRO certification in designing and building AI products and services.
AI + Design
How we work
Mindset
Section 01
Mindset
From giving commands to building a productive collaboration.
The real shift
AI does not improve design by itself. Better context changes how we work with it.
The opportunity is not more AI. It is better collaboration, grounded in evidence, design intent, and human judgment.
- Outcome
- Evidence
- History
- Constraints
- AI
Mindset
The best AI partner is not a tool or a know-it-all.
Too little agency gives us automation. Too much authority gives us false confidence.
- 01
Command
“Do exactly this.”
Too little agency gives us automation.
- 02
Collaborate
“Work through this with me.”
Ask, challenge, explore, refine.
- 03
Oracle
“Tell me the answer.”
Too much authority gives us false confidence.
The principle
Give AI enough agency to contribute, but not enough authority to own the decision.
AI + Design
How we work
Designing collaboration
Section 02
Designing collaboration
Giving AI the context it needs and carrying that understanding through the work.
Context
Context is what makes working with AI work.
Without it, AI can only be generically useful. With it, the work can become specifically useful.
- 01
Outcome
What needs to change?
- 02
Evidence
What do we know, and from where?
- 03
Constraints
What rules and limits matter?
- 04
History
What have we already tried or learned?
- 05
Current moment
What matters now?
Context is not one giant prompt. It is the shared understanding that accumulates through the work.
Workflow
With context, AI becomes part of the work, not a side task.
The workflows that stick attach to moments the team already has. They shorten a cycle we already run. They do not create a new “AI process.”
- 01
Project kickoff
Frame
- Human
- Outcome and success
- AI
- Interviews and challenges
- 02
Research readout
Synthesize
- Human
- Tradeoffs and guardrails
- AI
- Structures signals and logic
- 03
UX/UI design artifacts
Prototype
- Human
- What to learn
- AI
- Vibe-coded prototype and branches
- 04
User test or internal feedback
Evaluate
- Human
- Interprets evidence
- AI
- Synthesizes and compares
- 05
Retro and handoff
Capture + iterate
- Human
- Decision and direction
- AI
- Captures and updates
AI does not replace the process. It reduces friction between thinking, making, testing, and learning, and the workflow gets faster because context travels with the work.
Workflow / Synthesize
Context needs real evidence, not just conversation.
Trust comes from checking the model’s work, not from how confident it sounds.
Trusted inputs
- Research
- Product and business
- Design decisions
- Metrics and constraints
Before anyone relies on the synthesis
- 01
Separate the layers
Facts, inferences, and recommendations are kept apart.
- 02
Check the work
Claims, sources, and context are verified.
- 03
Let people decide
The tradeoffs stay with the team.
Missing evidence becomes a research question, not a generated answer.
Workflow / Prototype
Context tells AI what “good” looks like.
AI is already strong at behavior. It is much weaker when quality lives only in our heads.
Where it is strong
- Fast learning
- State logic
- Branches
- Edge cases
- Coded flows
Where it struggles
- 01
System fidelity
Needs precise components and implementation rules.
- 02
Visual polish
Taste is harder to encode than behavior.
- 03
Cross-platform consistency
Needs clear variants, constraints, and current documentation.
- 04
Missing intent
AI cannot recover knowledge we never captured.
Fast prototypes should give us better questions, not just faster output.
AI + Design
How we work
Building capability
Section 03
Building capability
Turning experiments into systems, products, and habits that can be trusted and repeated.
Scale
When a whole team uses AI, context needs a system.
Our design knowledge was made for people who know where to look. AI needs reliable access.
Today
- Figma
- Zeroheight and docs
- Slack
- Meetings
- People and SMEs
Overlapping, undocumented, and differently interpreted.
What AI needs
- 01
Scope
What matters.
- 02
Source
Where it came from.
- 03
Structure
How it connects.
- 04
Freshness
How it stays current.
- 05
Ownership
Who owns it.
Scale / Spec-driven development
Spec-driven development turns design goals into context we can reuse.
The goal is a reliable connection to the sources we already trust.
Inputs, intent, outputs, oversight
- 01
Trusted sources
Figma, docs, and product specs.
- 02
Structure
Rationale, patterns, and constraints.
- 03
AI workflow
Prototype, document, and implement.
- 04
Evaluate
Compare the result to the intent.
- 05
Refine
Test, learn, update, and repeat.
The model gets better when the system around it gets better.
Scale / Maturity
We have moved past experiments when results can be repeated.
A demo becomes a capability when it can be trusted, reused, and improved.
- 01
Embedded
AI supports recurring design work.
- 02
Grounded
The team reuses trusted context.
- 03
Intent holds
Design intent survives implementation.
- 04
Oversight
The team knows where AI can act alone.
- 05
Measured
Results are checked and gaps improve the system.
Experimentation asks “Can we?” Maturity asks “Can we do it again, with confidence?”
Product
Our product runs on context too.
A useful experience understands the person, the moment, and the rules.
- 01
User context
Identity, preferences, history, and accessibility.
- 02
Moment context
Intent, device, location, and current task.
- 03
System context
Capabilities, policies, access, and guardrails.
System behavior
- Experience
- Relevant
- Timely
- Explainable
- Safe
- Adaptable
Poor context does not just weaken the answer. It creates the wrong experience.
AI + Design
How we work
Practice
The new default
Design the context, not just the output.
AI expands what we can explore. Humans remain responsible for what deserves to exist.
- 01
Bring the outcome.
- 02
Build the context.
- 03
Invite challenge.
- 04
Verify the evidence.
- 05
Prototype to learn.
- 06
Capture what changed.
Start the conversation
- 01Before suggesting a solution, what do you need to know?
- 02Which of my assumptions might be wrong?
- 03What is evidence, and what is inference?
- 04What should we prototype first to learn?
- 05What missing context could change the answer?
What context would change the conversation?
If any of this raises a question for your team, I would like to hear it.