Startups and small teams have always valued speed, but AI can change where that speed comes from.

Instead of compressing traditional UX phases, AI allows us to skip straight to solutions and let learning happen through refinement.

In practice, this means moving away from drawn-out discovery and ideation phases and towards a flow where designers create something real almost immediately, then shape it with others until it earns the right to be polished and shipped.

The goal is still the same as classic lean UX: reduce risk, learn fast, and build the right thing. The difference is that AI dramatically lowers the cost of starting.

1. Frame the problem

0-1 day

Goal

Understand enough to begin the solution process

Approach

AI plays a practical role here by helping the team align quickly around what is known, what is assumed, and what is still unclear.

Designers can use AI to:

• Synthesise existing research, analytics, and support tickets into clear themes
• Combine inputs from different stakeholders into a single, neutral view
• Rephrase vague or conflicting problem statements until they are simple and concrete
• Surface gaps, assumptions, and contradictions early

At minimum, the team needs to be able to answer:

• Who is this for?
• What problem are we trying to solve?
• Why does it matter to users and the business?
• What constraints do we need to respect?

This understanding can come from existing research, analytics, support tickets, stakeholder conversations, or prior knowledge. The key is not where it comes from, but that it is explicit.

If you cannot explain the problem simply, AI will not help you solve it.

Output

A short problem statement and a clear set of assumptions and requirements that will shape the first solution.

2. Generate a starting solution

0-1 day

Goal
Create something real as fast as possible. The purpose is momentum, not correctness.

Approach
Use AI tools to draft a few prototypes directly from a problem statement. Treat the output as an initial hypothesis. Run a few iterations until the basic flow or UI is working, and covers all the initial requirements.

Claude.ai is excellent for generating iterative prototypes that can be easily shared.

In terms of ‘going wide’, this can still be part of the process. Any ideas around a different approach or model can simply be fed into the AI, and the results shared and evaluated. AI shouldn’t dictate the solution, it should just assist in getting there quicker.

Output
An initial prototype that can be shared and discussed.

3. Refine with the team

1-2 days

Goal
Turn an individual starting point into a shared understanding.

Approach
Review the solution with SMEs, engineers, product partners, and stakeholders. Focus discussion on:

• What feels right or wrong for users
• What is missing or unnecessary
• Where the flow breaks down or overcomplicates things

AI can help here by supporting alignment and synthesis. For example, AI can be used to:

• Capture and summarise feedback from workshops
• Group comments into clear themes
• Rephrase conflicting feedback into neutral problem statements
• Turn discussion points into clear, actionable change requests

This helps the team move from opinions to decisions more quickly, without losing nuance.

Output
A list of improvements, changes, and omissions.

4. Refine structure and interaction

1-2 days

Goal
Make the solution make sense before making it look good.

Approach
Take the first round of team feedback and feed it back into the AI agent to synthesise it into actionable insights - these can then be used to refine the prototype. Inputs can be specific (for example, “this step feels unnecessary, please combine it with the previous step”) or broad (“this flow feels too complex”).

AI helps here by making iteration cheap. Designers can quickly explore alternatives without redrawing everything by hand, which encourages comparison rather than commitment.

Typical refinement at this stage includes:

• Simplifying flows by removing or merging steps
• Clarifying information hierarchy and order
• Improving transitions between screens or states
• Making decisions more obvious and reducing cognitive load
• Exploring different interaction patterns where there is uncertainty

Designers guide this process by judging what improves understanding and what introduces noise. The goal is not to generate endless variations, but to converge on a flow that feels coherent.

Visual fidelity should remain restrained. This keeps feedback focused on structure and interaction, rather than colour, spacing, or brand.

Output
A clear, testable flow or prototype.

5. Validate quickly and selectively

2-4 days

Goal
De-risk the biggest assumptions with minimal overhead.

Approach
Run lightweight usability testing or structured internal reviews. Focus on:

• Task success
• Comprehension
• Decision confidence

AI can support this phase by reducing the effort needed to analyse and act on feedback, rather than replacing direct user input. For example, AI can be used to:

• Summarise usability session notes, recordings, or transcripts into clear findings
• Identify recurring issues and patterns across multiple sessions
• Map observed issues back to original assumptions or requirements
• Help prioritise problems based on frequency, severity, or impact
• Suggest potential fixes or alternatives to explore in the next iteration

Capture feedback directly in the design artefact and prioritise fixes by impact.

Output
An updated solution with major risks addressed.

6. Apply the design system and ship

1-2 days

Goal
Turn a validated solution into a consistent, shippable product.

Approach
AI can extend the designer’s role beyond static design files and into working product code.

In practice, designers can:

• Access the codebase directly to understand how components are implemented
• Use tools like Cursor to explore, modify, or extend existing components safely
• Prototype interactions and edge cases using real components instead of mocks
• Validate behaviour, states, and transitions in the actual product environment
• Hand over working examples or partial implementations to engineers, rather than abstract specifications

AI may also provide support by:

• Answering questions about design system usage or component variants
• Helping spot obvious inconsistencies or missing states
• Generating quick checklists for QA or handoff

In some cases, the entire prototyping and testing loop can happen in real code, inside the product itself. This reduces translation loss between design and build, and shortens the path from validation to release.

However, the responsibility for quality does not shift. Consistency, accessibility, and build quality still depend on design judgement and close collaboration with engineering.

Output
A signed-off build, supported by working examples in code, and a short, prioritised list of follow-up improvements.

What designers do differently in this model

In this flow, the designer’s value shifts towards:

• Framing problems clearly enough for AI to be useful, and assumptions explicit enough to test
• Synthesising inputs from users, stakeholders, and systems into shared understanding
• Recognising good solutions among many plausible ones, and knowing when to converge
• Using AI to iterate quickly across structure, interaction, and behaviour, not just visuals
• Working closer to the product itself, including prototypes and experiments in real code
• Facilitating alignment and decision-making across design, product, and engineering
• Knowing when something is good enough to move forward, and when to slow down to protect quality

This reflects a broader evolution in UX maturity: away from producing artefacts in isolation, and towards shaping outcomes across the whole product lifecycle.

Product design with AI is less about perfect process and more about momentum. By starting with solutions, refining them collaboratively, and applying craft late, teams can move faster without sacrificing quality.

Crucially, this approach also creates space to invest in interaction, motion, and detail — areas that are often deprioritised in the name of speed, despite their impact on clarity and experience.as.