Ask ten designers what they think about AI and you'll get ten different answers — ranging from "it's completely changed how I work" to "I tried it once and went back to doing things properly." Both responses are understandable. AI tools have arrived unevenly, and the hype cycle has made it genuinely hard to separate the signal from the noise.
After spending the better part of the past year integrating AI tools into real client work — not demos, not side projects, but actual live websites for paying clients — we've developed a clearer picture of what's actually shifting. It's less dramatic than the headlines suggest, and more significant than the sceptics admit.
The Shift Happening Now
The most important change isn't that AI is doing design. It's that the economics of exploration have collapsed. Where a designer once had to choose between exploring three layout directions or one, they can now explore fifteen in the same time. That changes the nature of the job fundamentally.
Historically, a lot of design decisions were made under time pressure. You chose a direction early, refined it, and delivered it — not necessarily because it was the best option, but because revisiting the brief at hour forty wasn't viable. AI removes that constraint in certain parts of the workflow.
What we're seeing in practice is a compression of the divergent phase of design. The period of generating options, exploring visual directions, and testing copy variants now takes a fraction of the time it used to. The convergent phase — deciding what's actually right for this brand, this audience, this moment — remains stubbornly human.
What AI Handles Well
There are three categories where AI tools have become genuinely indispensable in our workflow.
Layout Exploration
Framer AI and to a lesser extent Figma AI can generate credible layout structures from a brief in seconds. Not finished designs — but structural scaffolding that gives you something real to react to. The value isn't in using what's generated; it's in using it as a starting point for a critique that would have otherwise taken hours to set up.
Copy Variants
Writing five versions of a hero headline used to mean five minutes of staring at a blank document. With Claude or ChatGPT, you can generate twenty headline variants in under a minute, then spend your actual time making the editorial judgment about which direction works. The creative decision is still yours — the clerical work of generating options is delegated.
Image Generation and Visual Concepting
Midjourney has genuinely changed how we present visual direction to clients in early conversations. Rather than relying on stock photography or rough sketches, we can generate mood-aligned imagery that communicates a visual language quickly. It's not production-ready, but it bridges the gap between written brief and visual understanding faster than anything else.
The question was never "can AI design?" The question is "what does design mean when generation is free?" The answer involves more taste, more judgment, and more strategic thinking — not less.
What Still Needs a Human
The list of things AI cannot do well is, perhaps surprisingly, the list of things that make the difference between a good website and a forgettable one.
- Understanding what a brand actually stands for — beyond its stated values
- Reading a client's business context and making decisions that account for it
- Knowing when a design is "done" versus when it's merely finished
- Making the judgment call that the brief itself is wrong
- Building trust with clients and managing the emotional arc of a project
- Applying taste — the accumulated, largely inarticulable sense of what works
These are not edge cases. They are the core of what clients are actually paying for when they hire a design studio. A beautiful layout generated in seconds is worthless if it's solving the wrong problem. And AI, for all its fluency with patterns, has no way of knowing whether the problem it's been handed is the right one.
A useful test: if a task requires you to understand something that isn't in the brief — about the business, the culture, the competitive context, or the client's unstated anxieties — AI can't do it. If the task is about generating options within a defined space, AI is probably faster than you.
The New Designer Workflow
The workflows we've settled into look roughly like this: a project starts with a discovery and strategy phase that is entirely human. We're reading the business, asking questions, developing hypotheses about what the site needs to accomplish. AI might help with competitor research synthesis at this stage, but the thinking is ours.
As we move into visual design, AI enters more heavily. We'll use Midjourney to generate mood references, Claude to explore copy directions, and Framer AI to sketch structural layouts. This phase now takes roughly half the time it used to — which means we can either take on more projects or invest more depth in each one. We've chosen the latter.
Build and refinement remains largely manual. Framer is our platform of choice, and while Framer AI can scaffold a starting point, the detailed work — interactions, responsive behaviour, performance optimisation — is done by hand. This is where the craft lives, and it doesn't compress as easily.
Client review cycles have gotten shorter, partly because we can iterate faster, and partly because the quality of what we present in early rounds has improved. AI-assisted exploration means clients see more developed options earlier, which reduces the number of revision rounds.
Tools Leading the Change
If you're building a studio workflow around AI, these are the tools doing the real work right now:
- Framer AI — best for rapid layout generation and landing page scaffolding. Output quality is improving fast.
- Figma AI — more useful for iterating on existing designs than generating from scratch. The "make component" features are quietly excellent.
- Midjourney — still the best for photographic and painterly image generation. V6 is a significant step up from earlier versions.
- Adobe Firefly — slower but better for commercially safe imagery. Essential if your clients have IP concerns.
- Claude — our preferred tool for copy, brief analysis, and strategic thinking. Better at nuanced writing than ChatGPT.
- Perplexity — for research. Better than running raw searches because it synthesises rather than lists.
What This Means for Clients
For clients, the most immediate practical implication is speed. Work that used to take four weeks can sometimes be done in two — not because corners are being cut, but because the generative, exploratory phases of design are genuinely faster. The strategic and craft work takes the same time or more, because we're using the hours recovered from generation to invest more deeply in thinking.
The second implication is quality of exploration. Clients who engage early in the process now see richer sets of options than they would have previously. This makes early-stage decisions harder in a good way — you're not choosing between two directions because only two were viable to explore; you're choosing from eight because eight were genuinely feasible to develop.
The third — and this is important — is that the value of a designer's judgment has gone up, not down. When generation is cheap, curation becomes the scarce resource. A designer with strong taste and strategic instincts is more valuable now than they were when execution time dominated the cost structure. The profession isn't shrinking. It's shifting towards its highest and best use.
What AI is doing to web design is what spreadsheets did to accounting. The tool changed the work, made certain tasks trivially fast, and in doing so revealed more clearly what the human is actually for. That's not a threat to the profession. It's a clarification of it.