What AI actually does in a design project, and what it doesn't
AI now handles a specific, useful slice of design work. A mood board that once took a designer half a day to assemble by hand can be generated from a prompt in minutes. Layout variants, colour combinations and first-pass mockups can be produced in bulk, giving a client a wide spread of directions to react to early. First drafts of copy, icon sets and simple illustration can come out of a generative tool fast enough that the old excuse for a slow first meeting, "we need time to put some options together", no longer holds up.
None of that is brand strategy, and treating it as such is where buyers get into trouble. A tool can show you a hundred logo variations; it cannot tell you which one is right for a professional services firm trying to look established, because that judgement depends on the client, the market and the decision being made. The tool has no stake in the outcome and no memory of the conversation that shaped the brief.
What still needs a person
Four things stay firmly human, whatever the tooling looks like underneath:
- Judgement: choosing between options that are all technically competent, based on what the client actually needs.
- Brand strategy: deciding what the organisation should stand for and how that translates into visual choices, which requires understanding the business.
- Systems thinking: building a design system (the shared rules and components that keep a brand consistent as it grows) that holds together across a website, a set of documents and whatever gets built next year.
- Accountability: someone who can explain why a decision was made and stand behind it, which a generated output cannot do for itself.
A useful question when you are shown AI-generated work in a pitch is where in the process it was used and who made the calls that followed. An agency that generates thirty logo variants and picks the best one by eye is doing something different from an agency that generates thirty variants and hands you all thirty to choose from. The first is using the tool to speed up its own thinking; the second is asking the tool to do the thinking for it, then charging you for the privilege of sorting through the output.
The distinction matters because it is where the fee is either earned or not. If you want a closer look at what AI-produced work should look like before you accept it, Judging AI-Generated Design sets out what to check. If you are still deciding which route to buy through in the first place, Agency, Freelancer, Platform or AI: Which to Use for What compares the options directly.
What to ask before you sign
These are the questions worth putting to any agency, freelancer or platform before you commit.
-
Where does AI sit in the process, specifically
Ask which stages use AI tools and which are done by a person.
-
Who is accountable for the final result
A named person should be able to explain why a decision was made.
-
Is AI disclosure written into the proposal
If AI involvement isn't mentioned anywhere in writing, ask why, and treat a vague answer as information in itself.
-
What exactly do you own at the end
Source files, editable formats and IP assignment (the transfer of rights in the work to you) should be stated plainly.
-
Can you see the scope and quote in writing before work starts
A defined deliverable list and a written quote is the baseline.
-
Is there work you can actually check
Ask for something you can look at directly, such as a live site or a shipped project.
-
What happens if the brief changes
Ask how a scope change is handled and priced, since this is where vague arrangements tend to unravel.
-
Would they explain a design decision
If nobody on the other end can say why a choice was made, that's a reasonable reason to end the conversation.
Not every question will apply to every job. A small design refresh doesn't need the same scrutiny as a full identity and design system, but ownership and AI disclosure are worth asking about regardless of size. Use the Agency Scorecard on this site to score a shortlist against the same criteria.
Agency, freelancer, platform or AI: how they compare
The columns are the things worth checking before you commit. Match the row to the job you actually have.
| Route | Best for | Watch out for | What to check |
|---|---|---|---|
| Traditional agency | Brand identity work needing strategy, systems thinking and someone accountable for the result | Process can be opaque, and AI use is often undisclosed unless you ask directly | Ask to see past source files and IP assignment terms in writing before signing |
| Freelance designer | Smaller, well-defined jobs where you deal with one person throughout | Capacity is limited to one person, so timelines slip if something else comes up | Check what happens to ownership and files if they become unavailable mid-project |
| Design platform or marketplace | Quick turnaround on templated tasks like a logo variant or a simple mockup | Little visibility into who is actually doing the work or how much is templated | Read the platform's ownership terms closely. |
| AI tool used directly | Fast first-pass exploration: mood boards, layout variants, draft copy | No accountability if the output is wrong, and no judgement on whether it fits your brand | Ownership of AI-generated output is genuinely unsettled; check current Intellectual Property Office guidance |
| dsgn | A fixed-scope brief with a written quote before work starts, whatever mix of AI and human judgement is used | Scope is agreed upfront; if it changes, the quote is revisited before it does | You own the source files as standard. |
This is general guidance for comparing options, not a tested or audited comparison of named agencies. Use the Agency Scorecard on this site to run your own shortlist through the same questions.
Where to go next
Use the Agency Scorecard to compare your shortlisted candidates against the same criteria: process transparency, AI disclosure, ownership terms, pricing clarity.