How a Design Agency Uses AI: Where It Helps and Where It Doesn't

In a design agency AI works as an accelerator: it helps gather research faster, generate options for discussion, draft copy and handle routine operations. But strategy, the idea, the final design and accountability for the result still rest with people — and clients should understand where that line lies.

Why every client asks about AI

Clients ask about AI for two reasons. The first is an expectation that with neural networks everything should be faster and cheaper. The second is concern: was the logo generated in five minutes, is the result unique, could company data leak? Both questions are fair, and an honest answer matters more than marketing.

Where AI genuinely helps

Research and analysis

• a quick overview of the market, competitors and visual trends as a starting point;
• structuring interviews and briefs, spotting recurring themes;
• collecting the questions the audience asks — for site structure or content.

Everything AI produces at this stage gets checked: models can be confidently wrong and invent facts.

Finding a direction

• moodboards and visual sketches to discuss style with the client;
• quick variations of composition, colour and mood to rule out what doesn't fit earlier.

Copy

• drafts of page structures, headline options and descriptions;
• help adapting copy between Russian, Uzbek and English — always proofread by a native speaker.

Production and routine

• resizing and adapting layouts for different formats;
• retouching, background removal, image extension;
• generating supporting elements: textures, backgrounds, mockups;
• helping developers write and review code.

Where AI doesn't replace people

• Strategy and positioning. A model doesn't know your business, your constraints or anything that isn't written online.
• The idea. Models reproduce the average — what has been seen many times before. A recognisable brand is built on difference.
• Logo and brand mark. A generated logo creates risks around uniqueness and rights — see our article on the risks of AI-generated logos.
• Systems thinking. A brand identity is a set of rules that work across dozens of touchpoints. That's design, not image generation.
• Taste and accountability. A person decides that the result is good and right for the client.

AI shortens the path from brief to first options, but not the path from options to a decision. The most valuable thing in design is choice, not the number of images.

Rules for handling client data

AI can only be used safely with clear rules. Basic principles any agency should follow:

1. Confidential client materials — financial data, unreleased products, personal data — are not uploaded to public services.
2. Enterprise modes where inputs are not used for model training, or local solutions, are used instead.
3. Data is anonymised before going into AI: names, amounts and titles are replaced with placeholders.
4. The client knows where and how AI is used in their project and can restrict it.
5. Everything generated is reviewed by a person: facts, image rights, brand fit.

More on the risks in our article is it safe to upload documents to ChatGPT.

What a project with AI looks like, stage by stage

To make the line between the model's work and people's work clear, it helps to break a typical project — say, a website or a brand identity — into stages:

1. Brief. People lead the conversation with the client. AI can help structure the meeting transcript and list open questions — without uploading confidential details to public services.
2. Research. A model speeds up the initial overview, but a strategist formulates the conclusions and checks the sources.
3. Concept. Moodboards may include generated sketches, but the idea and the reasoning are the work of the designer and art director.
4. Design. Layouts, grids, typography and the logo are hands-on design work. AI helps with supporting tasks: photo processing, background options.
5. Production. This is where savings are most visible: format adaptations, resizes, draft translations, help with code.
6. Quality control. Every piece of output is checked by a person: copy by an editor and a native speaker, images for artefacts and rights, code for bugs and security.

Common mistakes when using AI in design

• Treating the first generated option as the solution. Without design logic, the result looks impressive but doesn't serve the task.
• Not checking details. Distorted hands, illegible lettering, stray background elements — artefacts that undermine trust in the brand.
• Using machine-generated Uzbek without proofreading. Language errors are obvious to the audience straight away and come across as disrespectful.
• Mixing styles. Images from different generators and prompts without shared rules break the visual system.
• Uploading everything to services. Contracts, customer databases, unreleased products — materials that don't belong in public tools.
• Promising clients “many times faster and cheaper”. Individual operations speed up, but approvals, thinking and review remain.

How clients can use AI on their side

Models can help the client too — if used as an assistant rather than a substitute for the contractor:

• draft a brief and a list of questions for the agency;
• collect competitor examples and describe what you like or dislike about them;
• prepare draft product copy that an editor will then refine;
• quickly sketch ideas for discussion — understanding that this is a starting point, not finished design.

The main rule stays the same: don't upload confidential company or customer data to public services.

Classifying materials: what can be given to AI

A simple three-level scheme saves the team from deciding afresh each time whether a model can be used:

• Public materials — published copy, publicly available market information, your own drafts without client data. Can be used in approved tools.
• Internal materials — briefs, working layouts, project correspondence. Only in enterprise modes with no training on inputs, and after anonymisation.
• Confidential materials — personal data, financial information, unreleased products, documents under a non-disclosure agreement. Never uploaded to public services; if needed, local solutions only.

This classification is worth agreeing with the client at the start of a project, especially if they work in banking, healthcare or with government data.

What this changes for the client

• Faster first iterations: you see directions earlier and can discuss them.
• More time for what matters: strategy, detailing and quality rather than routine resizing.
• The price doesn't necessarily drop: AI saves hours on routine, not on thinking, and a project's value lies in its decisions.
• Experience is needed to tell good from plausible: generating a hundred options is easy; choosing the right one is a professional's job.

Copyright and transparency

A separate issue is rights to the result. The legal status of AI-generated images is still taking shape in many countries, and the terms of different services vary: some allow commercial use, others restrict it or don't guarantee freedom from third-party claims. So a sensible practice looks like this:

• key brand elements — logo, brand mark, original illustrations — are created by designers so the rights can be cleanly transferred to the client;
• generated materials are used as supporting assets or after substantial rework;
• for commercial work, tools whose terms explicitly allow such use are chosen;
• the contract states who owns the rights to final files and source files;
• the client is told which elements were created with AI.

Transparency works both ways here: the client understands what they're paying for, and the agency doesn't put its reputation at risk if questions about an image's origin come up later — for example, when registering a trademark or entering a foreign market.

Questions to ask an agency

1. Do you use AI on my project, and at which stages?
2. Which services will my materials go into, and are they used for training?
3. Who owns the rights to the result if it contains generated elements?
4. How do you check the uniqueness of key brand elements?

A good agency will answer these questions directly. If you need a team that uses AI as a tool rather than a substitute for design, Dasta Creative works through design support and project-based engagements.