Design work rarely happens inside one application anymore. A typical project can involve a Figma file, competitor research, copywriting, image references, UX analysis, presentation material, and a dozen browser tabs before the actual interface is even finished.
The problem is not a lack of AI tools. There are too many of them.
For designers, the useful question is different: Which AI tools actually deserve a permanent place in your workflow?
Based on current capabilities and practical design workflows, this article looks at six tools that cover different parts of the process: interface exploration, visual creation, research, UX thinking, content, and prototyping. The goal is not to let AI design everything for you. It is to reduce repetitive work and give you more room for decisions that require actual design judgment.
1. Figma AI Keeps Design Exploration Inside the Canvas
For UI and UX designers, Figma remains the natural centre of the workflow, and its AI features are increasingly becoming part of that workflow rather than a separate add-on.
Figma’s current AI direction goes beyond generating a quick screen. Its agent can generate and remix designs, edit images, search files and help automate repetitive work. Figma Make can also turn ideas and existing designs into functional prototypes and interactive web experiences. (Figma)
That matters because context is often more valuable than raw generation speed.
Use Figma AI for first drafts, not final decisions
Figma’s First Draft functionality can transform a description into editable wireframes or designs. As of May 2026, Figma’s agent became the new entry point for this workflow and is being rolled out as a beta feature. (Figma Help Center)
A useful workflow is:
- Describe the product and target user.
- Generate a rough structure.
- Inspect the information hierarchy.
- Replace generic components with your design system.
- Rework spacing, typography and interaction states manually.
- Test the resulting flow with realistic content.
The important part is step four.
AI-generated interfaces often look convincing because they contain familiar patterns. That does not mean the hierarchy is appropriate for the product. A designer still needs to ask whether the CTA has the right priority, whether the navigation reflects the user’s mental model, and whether important edge cases have been considered.
Figma Make is useful when static screens stop being enough
A static frame can hide serious interaction problems.
A pricing page may look excellent until you test switching plans. A dashboard may appear balanced until filters, empty states and error messages are introduced. A mobile onboarding flow may look polished until you actually click through it.
Figma Make is designed for this middle ground. It can turn prompts and existing Figma designs into functional prototypes, web apps and interactive UI. (Figma)
That makes it particularly useful during design reviews.
Instead of saying, “Imagine that this interaction happens,” you can show the interaction.
Best use: UI exploration, rapid prototyping, interaction testing and design-system-aware iteration.
Watch out for: accepting generated UI simply because it looks polished.
2. ChatGPT Works Best as Your UX Thinking Partner
A designer does not only produce screens.
You also need to understand a brief, identify assumptions, structure content, review user flows, analyse competitors, write microcopy and explain design decisions to clients or developers.
That is where ChatGPT can sit beside Figma.
Its current research capabilities can search the web, compare sources and produce structured, cited research. Deep research can go further by planning multi-step research and synthesising information from multiple sources. (OpenAI)
Use it before opening Figma
One of the most useful changes to a design workflow is moving some thinking earlier.
Instead of opening Figma and immediately creating a hero section, give the AI the brief first.
For example:
“We are designing a website for a B2B logistics company. The primary users are procurement managers and operations teams. Identify the five most important questions a first-time visitor needs answered before requesting a quote.”
That produces something more useful than asking for “a modern logistics website.”
You can then turn those questions into the information architecture.
The same approach works for UX research summaries, persona assumptions, onboarding flows, usability-test questions and stakeholder interviews.
Use image generation when visual exploration is the bottleneck
ChatGPT can also generate and refine images from natural-language prompts. You can request variations, change composition and explore different visual directions without manually producing every concept. (OpenAI)
For designers, this is particularly useful at the concept stage.
You might need:
- A hero image showing a specific product context
- Editorial imagery for a website concept
- Visual references for a moodboard
- Product-scene variations
- Illustration directions
- Background concepts
- Social campaign concepts
The best workflow is not “generate an image and publish it.”
Generate several directions, identify what works, then art-direct the final result.
Best use: UX reasoning, research, copy, design critique, visual ideation and client communication.
Watch out for: treating AI-generated research or recommendations as automatically correct. Even OpenAI notes that deep research outputs should be reviewed and that models can still make incorrect inferences. (OpenAI)
3. Adobe Firefly Handles Visual Production Without Leaving the Adobe Ecosystem
There is a difference between generating a nice picture and producing a usable creative asset.
Designers frequently need to remove an object, extend a background, create variations, produce a product scene or generate supporting imagery while keeping the asset compatible with a broader Adobe workflow.
That is where Firefly becomes useful.
Adobe’s current Firefly environment brings image and video generation, editing, asset organisation and project workflows into a single creative environment. (Adobe Help Center)
Use Firefly for production problems
Imagine you have a product photograph that is almost perfect.
The composition works, but the left side needs more space for headline text.
Instead of searching for another photograph, you can use generative tools to extend the scene.
Other practical examples include:
- Removing distracting objects
- Creating alternate backgrounds
- Generating visual variations
- Extending an image for different aspect ratios
- Creating product environments
- Producing campaign concepts
- Exploring image and video directions
Firefly’s image-generation tools can produce different visual styles, while image-to-image workflows can use an existing visual reference as a starting point. (Adobe)
Commercial work requires a different level of attention
This is one area where designers should read the terms instead of assuming every AI image has identical usage rights.
Adobe says its own Firefly models are trained on licensed and public-domain content and positions their outputs as commercially safe. Adobe also states that it does not train Firefly models on customer content. Partner models available through Firefly can have different terms. (Adobe)
That distinction matters when you are creating assets for paying clients.
If a client asks, “Can we legally use this in our advertising campaign?”, the answer should depend on the model, asset source, applicable terms and local law, not simply on the fact that an image was generated by AI.
Best use: production imagery, image editing, visual variations and Adobe-based creative workflows.
Watch out for: assuming every model available inside a platform has identical commercial-use conditions.
4. Perplexity Is Useful When Your Design Decision Needs Evidence
A designer researching a new project often needs answers to questions such as:
- What competitors are doing?
- What terminology does the industry use?
- What features are common?
- What are users complaining about?
- What regulations affect the product?
- What information should appear on a product page?
Traditional search can answer these questions, but it can also create a messy collection of tabs.
Perplexity’s Pro Search is designed to synthesise information from multiple sources into a more detailed answer. Its current documentation describes it as a research workflow for complex questions, with access to multiple AI models depending on the plan. (Perplexity AI)
Use research tools to support design decisions
Suppose you’re redesigning a healthcare website.
Rather than asking:
“What should a healthcare website look like?”
Ask:
“Analyse ten leading private healthcare websites and identify recurring patterns in appointment booking, doctor discovery, service navigation and trust signals. Separate observed patterns from your interpretation.”
That second question is far more useful.
You can turn the result into a research document containing:
| Research question | Design output |
|---|---|
| What do competitors prioritise? | Sitemap |
| What information appears repeatedly? | Content hierarchy |
| Where do booking journeys start? | User flow |
| What trust signals are common? | UI/content requirements |
| Where do competitors create friction? | Opportunity areas |
Always inspect the source behind the answer
Research assistants are excellent at reducing the time required to find information, but they are not substitutes for checking evidence.
Open the source.
Read the relevant section.
Ask whether the claim actually supports the design decision.
This is particularly important when research affects healthcare, finance, legal services, accessibility or other regulated areas.
Best use: competitor research, industry research, source discovery and evidence gathering.
Watch out for: copying a competitor’s interface simply because the research identified it as common.
5. Claude Is Worth Keeping Open for Design Critique and Long Context
Claude is particularly useful when the design problem involves a large amount of information rather than a single visual task.
A designer might have a 30-page client brief, user research notes, existing website content, brand guidelines and a sitemap. Reading everything manually before making a decision takes time.
Claude can help structure that material and turn it into design-relevant questions.
Anthropic’s current Claude Design product also moves Claude further into visual work, allowing users to create designs, prototypes, slides and one-pagers through conversation and direct refinement. (Anthropic)
Give Claude the messy material
This is where many designers get better results than they do from a generic “design me a website” prompt.
Give it the actual material:
- Client brief
- Existing website copy
- Brand guidelines
- Customer interview notes
- Product documentation
- Sitemap
- Competitor observations
Then ask it to identify contradictions.
For example:
“Review this website brief from a UX perspective. Identify unclear requirements, conflicting priorities, missing user states and assumptions that should be validated before design starts.”
That is a much stronger use of AI than asking it to produce another generic landing-page concept.
Use AI as a critic, not just a generator
A useful critique prompt can ask for problems rather than compliments:
“Review this onboarding flow as a first-time user. Identify where I might hesitate, what information is missing, and which steps could create unnecessary cognitive load. Do not redesign it yet.”
This changes the role of AI.
You are no longer asking, “What should I make?”
You are asking, “What might I have missed?”
That distinction matters because professional design is often about finding problems before users do.
Best use: design critique, requirements analysis, long documents, UX reasoning and visual exploration.
Watch out for: allowing the AI to replace actual usability testing with people.
6. Gemini Adds Research, Visual Ideation and Google Workspace Context
Gemini is useful when your design work already lives heavily inside the Google ecosystem.
The current Gemini app in India includes image generation and editing, Deep Research, Canvas and Gems, alongside integration with Google services depending on the plan. (Gemini)
Its current image-generation experience also supports creating and editing visuals, including changing composition, style and image elements. (Gemini)
Use Gemini when research and visual work overlap
A practical design scenario might look like this:
You are designing a campaign for a school, travel company or consumer brand.
You need to:
- Research the target market.
- Understand the audience.
- Develop campaign concepts.
- Generate visual directions.
- Prepare supporting documents.
- Refine the presentation.
Gemini can be useful across several of those steps, particularly if the source material already sits in Google Drive or the output needs to move into Google Docs or Slides.
Its Canvas environment also supports creating and working with structured content, while Gemini can generate files such as Docs, Sheets and Slides from prompts. (blog.google)
Use image generation for variations, not final art direction
Gemini’s current image tools can modify references and change visual characteristics such as lighting, camera angle, focus and style. (Gemini)
That makes it useful when you already know roughly what you want.
For example:
“Keep the composition and subject, but create three visual directions: premium editorial, youthful technology brand, and minimal Scandinavian.”
You can then compare the directions before deciding what deserves further development.
Best use: research, visual exploration, presentation preparation and Google Workspace workflows.
Watch out for: assuming a generated visual is ready for production without checking brand consistency, typography, factual accuracy and licensing requirements.
How These Six Tools Fit Into One Design Workflow
The biggest mistake is treating six AI tools as six separate destinations.
A better approach is to give each tool a job.
| Design task | Tool to keep open |
|---|---|
| UX thinking and brief analysis | ChatGPT |
| Interface exploration | Figma AI |
| Functional prototypes | Figma Make |
| Visual production | Adobe Firefly |
| Evidence and competitor research | Perplexity |
| Long-form critique and requirements | Claude |
| Google-based research and visual ideation | Gemini |
You do not need to use all six on every project.
A typical website project could look like this:
Step 1: Understand the problem
Use ChatGPT or Claude to break down the brief, identify assumptions and formulate research questions.
Step 2: Research the market
Use Perplexity or Gemini to investigate competitors, terminology, user expectations and industry patterns.
Step 3: Build the information architecture
Return to ChatGPT or Claude and turn research findings into a sitemap, content hierarchy and user flows.
Step 4: Explore the interface
Use Figma AI to create several structural directions rather than polishing the first generated screen.
Step 5: Create visual assets
Use Firefly or Gemini to explore imagery, backgrounds, product scenes and campaign directions.
Step 6: Prototype behaviour
Use Figma Make to test interactions that are difficult to communicate through static frames.
Step 7: Critique
Put the resulting design back through an AI critique workflow. Ask specifically about hierarchy, accessibility, content clarity, interaction states and missing edge cases.
Step 8: Make the human decision
This is where the designer remains responsible.
AI can produce options.
It can identify patterns.
It can challenge assumptions.
It cannot automatically know which trade-off is right for your particular client, product, users and business constraints.
FAQs
Which AI tool is most useful for UI/UX designers?
Figma AI is particularly useful for designers who already work in Figma because its AI capabilities operate close to the actual design canvas. ChatGPT and Claude are useful alongside it for research, UX reasoning and critique.
Should designers use AI to generate complete website designs?
AI-generated designs are useful as starting points and exploration material. They should not automatically become the final interface. Information architecture, accessibility, content hierarchy, interaction states and product constraints still need human review.
Can AI replace UX research?
No. AI can accelerate desk research, organise research material and help formulate questions, but it does not replace interviews, usability testing, observation or direct feedback from real users.
Which AI tool is best for generating design images?
The answer depends on the job. Firefly is useful for Adobe-oriented production workflows and offers its own Firefly models with a commercial-safety position from Adobe. Gemini and ChatGPT are also useful for visual exploration and iterative image generation. (Adobe)
Should designers keep multiple AI tools open?
Yes, when each tool has a clear role. Keeping six tools open only because they are popular creates distraction. Keeping two or three that cover your actual workflow can save considerably more time.
What is the biggest mistake designers make with AI?
Treating generation speed as design quality. A screen can be produced in seconds and still solve the wrong problem. Good design work starts with understanding the user, context and constraints.
Conclusion
The most useful AI setup for a designer is not the one with the largest number of subscriptions. It is the one where every tool has a specific job.
Figma AI can shorten the distance between an idea and an editable interface. ChatGPT and Claude can help structure briefs, question assumptions and critique decisions. Perplexity and Gemini can reduce research time. Firefly can speed up visual production and exploration. Figma Make can expose interaction problems earlier through functional prototypes.
The common thread is context.
Give AI a vague prompt and you will usually get a plausible answer. Give it a real brief, real constraints, real research and a clear task, and it becomes much more useful.
For experienced designers, that is probably the most practical way to think about AI: not as a replacement for design skill, but as a set of assistants that remove some of the slower work around it.