5 levels of AI maturity in design
AI's role in design is expanding from completing small production tasks to participating in research, exploration, evaluation, and execution.
This five-level model describes that progression. Each level reflects a different working relationship between the designer and AI, with specific capabilities, responsibilities, and examples.
The five levels
| Level | AI's role |
|---|---|
| Level 1: Mechanical | A fast pair of hands |
| Level 2: Generative | An ideation partner |
| Level 3: Evaluated | A sparring partner |
| Level 4: Intentional | Executes toward a specific experiential quality |
| Level 5: Autonomous | An independent designer across the full process |
Level 1: Mechanical
At the mechanical level, AI is a fast pair of hands. It completes bounded tasks with clear instructions and limited product context.
Examples include:
- Writing error messages, button labels, and placeholder copy
- Removing image backgrounds
- Renaming and organizing layers
- Documenting a component
- Creating dark mode or responsive variants
- Resizing assets for different platforms
- Summarizing research transcripts
- Converting rough notes into structured documentation
- Checking a design file for inconsistent spacing or missing states
The designer defines every task and reviews every output. A request might be as specific as, “Write five versions of this empty-state message using our voice guidelines,” or “Create mobile variants of these three desktop cards.”
AI speeds up the workflow while leaving its structure unchanged. The designer still decides what needs to happen, breaks the work into steps, and assembles the final result.
Level 2: Generative
At the generative level, AI becomes an ideation partner. It receives a problem and produces a landscape of possible directions.
Imagine a team designing an interface for managing multiple agents. AI might generate four interaction models:
- A mission-control dashboard showing every active agent
- A dock for switching between agents
- A spotlight-style command interface
- An inbox that organizes requests, updates, and approvals
For each direction, AI can research comparable patterns, explain the interaction model, identify likely advantages, and create an initial prototype.
Other examples include:
- Exploring different onboarding structures for a complex product
- Generating navigation models for a growing application
- Proposing ways to communicate uncertainty in an AI-generated answer
- Creating several checkout flows for different customer behaviors
- Producing visual directions based on a brand strategy
- Exploring how a feature might work on mobile, desktop, and voice interfaces
- Generating multiple information hierarchies for the same dataset
- Suggesting different permission models for a collaborative product
The designer supplies the problem, constraints, customer context, and product goals. AI expands the solution space.
Level 3: Evaluated
At the evaluated level, AI becomes a sparring partner. It generates directions, gathers evidence, and challenges early assumptions.
Consider a study about how people manage multiple agents. AI runs a series of unmoderated sessions and notices a recurring behavior: participants treat the agent as a command executor, even though the product presents it as a collaborator.
AI proposes several responses:
- Change the study to investigate expectations of agency and control
- Test language that makes collaboration more explicit
- Compare a conversational interface with a task-management interface
- Introduce moments where the agent asks for clarification
- Examine whether users want collaboration at all
The designer or researcher decides whether the study should change direction.
Other examples include:
- Testing three onboarding concepts and identifying where users hesitate
- Comparing how novice and expert users interpret an AI recommendation
- Finding accessibility problems across generated interface variants
- Detecting that participants consistently misunderstand a key label
- Reviewing support tickets to identify gaps in an existing workflow
- Simulating edge cases before a concept reaches usability testing
- Comparing prototype behavior against a predefined design rubric
- Flagging when research participants do not represent the intended audience
- Debating whether a finding reflects a product issue, a study issue, or a recruitment issue
Level 3 requires an explicit evaluation system. The team defines what success means, which evidence matters, and which decisions AI may make during a study.
The human role includes writing the rubric, reviewing recruitment, setting boundaries for adaptation, and interpreting the findings.
Level 4: Intentional
At the intentional level, AI can execute toward a specific experiential quality.
This level concerns the details that shape how a product feels:
- The difference between optical and mathematical corner radii
- The speed and curve of an animation
- The rhythm of a multi-step interaction
- The amount of friction before a consequential action
- The visual weight of a warning
- The tone of an interruption
- The timing of progressive disclosure
- The density of information on a monitoring screen
- The way a product communicates confidence or uncertainty
- The transition from an empty state to an active workspace
For example, a team may want an approval interaction to feel deliberate without feeling slow. AI could generate several versions with different motion curves, confirmation patterns, delays, and visual treatments. It could test the options, compare behavioral responses, and refine the execution toward the desired quality.
A healthcare product might need an alert to feel urgent without creating panic. A financial product might need a confirmation step to communicate consequence without making routine actions exhausting. A creative tool might need an AI suggestion to feel inspiring without taking control away from the user.
These qualities are difficult to express as fixed specifications. Words such as “calm,” “confident,” “playful,” or “premium” can produce many technically valid interpretations.
The designer guides AI using references, critiques, lived experience, customer knowledge, and taste. Each execution clarifies what quality means in context. Once AI reaches the current standard, the designer notices new opportunities, raises the standard, and directs the next iteration.
This moving quality bar makes Level 4 an unsolved area. AI can already produce polished work. Consistently creating the appropriate emotional effect across a complete experience requires an evolving sense of what good looks and feels like.
Level 5: Autonomous
At the autonomous level, AI operates as an independent designer across the full process.
A designer or product lead might write:
From that brief, AI would:
- Research the users and their current workflows.
- Review existing product patterns and technical constraints.
- Identify the main problems and opportunities.
- Generate several product directions.
- Build testable prototypes.
- Recruit appropriate participants.
- Run studies and analyze the findings.
- Revise or eliminate weak directions.
- Select the strongest solution.
- Produce an annotated, production-ready specification.
- Hand the specification to a developer or implementation agent.
Other Level 5 scenarios might include:
- Redesigning a settings system after analyzing usage data and support requests
- Creating a new enterprise permission model from research through specification
- Developing an onboarding experience for a newly launched product
- Discovering a retention problem, testing several interventions, and preparing the strongest one for implementation
- Designing a cross-platform workflow across web, mobile, and voice
- Continuously improving a mature product through research, experimentation, and measured releases
Level 5 requires every preceding capability to work together reliably. Research quality, concept generation, evaluation, interaction design, visual execution, accessibility, edge cases, and documentation all become parts of one connected system.
Where most teams actually are
| Level | Status |
|---|---|
| Level 1: Mechanical | Already standard practice |
| Level 2: Generative | Where many teams operate today |
| Level 3: Evaluated | A practical stretch goal |
| Level 4: Intentional | An unsolved, moving target |
| Level 5: Autonomous | Beyond the current planning horizon |
Using the model
A team can operate at several levels at once. AI might handle production tasks at Level 1, concept exploration at Level 2, and a narrowly defined research program at Level 3.
The model is useful because it makes each working relationship concrete. For any AI-assisted design activity, teams can ask:
- What context does AI receive?
- What work can it complete independently?
- How is the output evaluated?
- Which decisions require approval?
- What evidence supports the final direction?
- Who owns the outcome?