has quickly become part of the workflow in , and web development.

It can reduce time spent on research, ideas, production and code. But the main change may be something else: as producing proposals becomes easier, assessing them becomes more important.

Use is increasing rapidly

In 2025, Figma surveyed 2,500 designers and developers about AI use. 78 percent said it increased efficiency, and 85 percent believed learning to work with AI would matter for future success.

Quality assessments were more cautious. 54 percent of designers believed AI improved their work's quality, compared with 67 percent of developers.

The figures show an important distinction: productivity and quality are different.

Graphic design: more directions in less time

In graphic design, AI can make exploring several alternatives early in the process easier.

Tools can support visual directions, image editing, text variations, simple mockups, format adaptations and repetitive production tasks.

Figma reported in 2024 that 59 percent of surveyed designers and developers used AI at work. Common uses included research, images, text, ideas and brainstorming.

The pace changes when ten directions can be explored in the time previously needed for two. But the options still need professional assessment. Typography, composition, brand, production requirements and message are not correct simply because a proposal looks finished.

AI produces alternatives. Design expertise determines which are worth developing.

UX: a useful assistant, a poor replacement for users

AI can contribute to UX long before a finished interface exists.

Nielsen Norman Group surveyed 841 UX professionals and described four common roles for generative AI: content editor, research assistant, idea partner and design assistant. Among AI users, 63 percent used it at least several times a week.

It can help with interview guides, survey questions, note summaries, microcopy, organising research and early user-flow or proposals.

The boundary is the user insight itself.

Nielsen Norman Group has also tested synthetic users: AI-generated people simulating research participants. Its conclusion was that these models can support preparation and hypotheses, but should not replace research with real users.

A language model can formulate a plausible answer. It does not have the experience, context or behaviour of the person who will use the product.

Development: the benefit is easier to measure

In web development, the effect is often more concrete.

Stack Overflow's 2025 Developer Survey found that 84 percent of respondents used or planned to use AI tools in development. Among professional developers, 51 percent used these tools daily.

A controlled Microsoft Research study asked developers to complete the same JavaScript task. The group using GitHub Copilot completed it 55.8 percent faster than the control group.

That is interesting, but it does not measure all development generally. The study concerned one defined task. Large digital products also need architecture, security, integrations, content and testing.

It nevertheless shows how much time AI can save when a problem is clearly defined.

'Almost right' has its own cost

Speed has a downside.

In the Stack Overflow survey, 46 percent of developers actively distrusted the accuracy of AI output, while 33 percent expressed trust. The most common frustration was solutions that were almost right but not quite, mentioned by 66 percent.

The same problem exists in design and content.

A result can look professional while containing incorrect facts, weak hierarchy, inconsistent details, accessibility problems or untested assumptions about users.

When production becomes faster, quality checks need to improve rather than weaken.

Expertise determines when a result can be used

AI can improve productivity even for less experienced users. Experience still matters for understanding the limits.

A designer can recognise more quickly why a composition or typographic approach fails. A UX designer can distinguish an interesting hypothesis from documented user insight. A developer can detect code that works in a demo but creates security or maintenance problems.

A study with Boston Consulting Group and Harvard Business School illustrates this. 758 consultants completed tasks resembling normal knowledge work. Within what the researchers called the technology's 'jagged frontier', participants worked over 25 percent faster and completed over 12 percent more tasks.

On a more complex task outside the model's strengths, AI users were 19 percentage points less likely to reach the correct answer.

The study concerned consultants, not designers. The principle is still relevant: a tool can perform very well on one task and surprisingly poorly on the next.

A robust workflow has several checkpoints

Professional AI use should be part of a normal professional process:

1. Define the problem, audience, content and constraints. 2. Use AI to explore alternatives and reduce repetitive work. 3. Assess the result against real requirements. 4. Test design, research and code in the right context. 5. Check accessibility, content, performance, security and consistency.

These checkpoints are the main difference between a robust workflow and simply prompting, generating and publishing.

Assess data and confidentiality

Professional AI use also concerns which data the tools can access.

Confidential client content, personal information and internal business information should not be entered into open consumer tools without clarifying data processing, access and responsibilities.

This is not an argument against AI. It requires the same risk assessment as other digital services.

The tool changes the division of work

Design tools have always changed how work is done. Digital layout software made production more efficient, Photoshop changed image work and Figma made collaboration and more accessible.

Generative AI goes further by proposing the text, image, layout or code itself.

It does not reduce the need for professional assessment. It moves more value towards understanding the problem, selecting, checking and editing.

AI can create a proposal in seconds. Knowing whether it is correct, relevant and safe to publish remains a separate skill.