Every design studio gets asked the same question now: how much of the work is AI? The honest answer for us is that AI touches most stages of a project and decides none of them. This is a first-person account of where it changed our process and where it didn't, described in general terms, without productivity numbers we can't prove.
How has AI changed the design process in 2026?
AI has changed the design process by making options cheap to produce, which moves the real work to choosing between them. A layout exploration, a set of headline variants or a component scaffold that once took a working session can now be drafted in the time it takes to describe it.
What didn't change is the order of the work. Our website design process still runs from discovery to messaging, structure, visual design, build, QA and launch, and each stage still ends with something you approve. AI sits inside those stages as a drafting tool. It doesn't merge them, and it doesn't skip the approvals, because the approvals are where a founder's knowledge of their buyers enters the project.
The side effect is subtle. When a polished draft costs almost nothing, it's easy to mistake polish for progress. A generated homepage can look finished on day one and say nothing specific about your product. So the main skill on a 2026 design team isn't prompting. It's knowing which of ten plausible options is right, and why.
Where does AI help in our design workflow?
AI helps most in our design workflow at the edges of each stage: gathering inputs before a decision and producing variants after one. It's least useful at the moment of the decision itself.
Stage | Where AI helps | Where people decide |
|---|---|---|
Discovery | Summarizing call notes, reviews and competitor pages | Which buyer and which problem the site is for |
Messaging and copy | Headline and CTA variants, tightening drafts | The main claim, the proof and what not to say |
Structure | Sitemap drafts and section order options | Which pages exist and what each must prove |
Visual design | Moodboards, layout exploration, image concepts | The brand direction and the final system |
Development | Component scaffolding, boilerplate, refactors | Architecture, performance and code review |
QA and launch | Checklists, typo and alt text sweeps | Sign-off on accessibility, states and content |
In discovery, AI is a good first reader. It can sort a pile of sales call notes into recurring objections, which gives us a better starting point for the questions in our website discovery phase. In development, it writes the repetitive parts of a Next.js or Framer build, so more of our time goes to the parts users feel: speed, states and interaction. Our guide to micro-interactions covers the kind of detail that time goes into.
Why do we still write the words first, even with AI?
We still write the words first because a layout can only arrange a message, and AI can't tell you which message is true for your company. It can draft a headline in seconds, but it drafts from patterns, and the pattern for a SaaS hero is the same for everyone.
The words come from what only you know: why customers switched, what they nearly bought instead, what a sales call keeps stumbling on. We use AI to produce variants once that argument exists, and to cut a 40-word sentence to 15. We don't use it to decide the argument. If you're weighing who should own your copy, our comparison of who writes website content explains the trade-offs between founders, agencies and specialist writers.
AI drafts, people decide, and the words come before the layout.
Writing first also makes AI tools better. A wireframe generator given a real message document returns a structure built around your proof. Given a one-line prompt, it returns the same hero, three features and testimonial slot it gives everyone else.
What does AI still get wrong in web design?
AI still gets the specific things wrong in web design: claims, edge cases, accessibility details and anything that depends on context it was never given. It's fluent, which makes these errors easy to miss.
- Claims and numbers. Generated copy invents plausible statistics and features. Every claim on a page needs a person who can point to where it came from.
- States. Generated layouts show the happy path. Empty, loading, error and long-content states still need designing, and our web app UI design guide explains why they matter.
- Accessibility. Contrast, focus order and labels often look fine in a draft and fail in use. We check them against WCAG, as in our accessible design basics.
- Sameness. Models converge on the average. Left alone, AI-assisted sites start to look like each other, which is the opposite of what a brand is for.
- Language and layout. Translated strings, longer languages and right-to-left text break generated layouts in ways a model won't flag. See designing for multiple languages.
What do we never hand to AI?
We never hand AI the decisions someone has to be accountable for: positioning, factual claims, brand direction, code that ships unreviewed and the final sign-off. These are the parts of a project you're paying for, and they're also the parts where a confident mistake costs the most.
Code is a good example. AI writes a lot of scaffolding for us, and every line of it is read and tested by the person who owns that part of the build before it merges. The same goes for QA: AI can sweep a site for missing alt text or broken links, but a person still runs the checks in our website QA testing guide on real devices before launch. It's also fair to ask any studio which AI tools will see your files and what those tools' data policies say.
For example, suppose a startup brief goes through an AI-assisted week
Hypothetical example: a seed-stage SaaS startup needs a new homepage. On day one, AI summarizes twenty pages of call notes into recurring objections, and a person picks the three that matter. The founder approves a one-page message: who it's for, the main claim, three proof points and one action.
AI then drafts a dozen headline variants and several section orders. The team discards most, rewrites the best two and designs the page around the strongest proof. In the build, AI scaffolds components, and a developer reviews and tests each one. Before launch, an automated sweep catches small issues, and a person checks contrast, focus and every claim. AI did a lot of typing. People made every decision that reached the page.
How should you judge a studio's use of AI in the design process?
Judge a studio's use of AI in the design process by how clearly it can show you where AI is used and who reviews the output, not by how fast it produces a first draft. Speed is easy to demonstrate. Judgment only shows up when you ask specific questions.
Ask for the map
Ask them to walk through a recent project and name each step that uses AI and each step where a person decides.
Find the reviewer
Ask who checks copy claims, accessibility and code before work reaches you, and how senior they are.
Ask about the words
Ask when copy gets written. If the answer is after the design, generated placeholder text is doing the strategy.
Check ownership and data
Confirm you own the files, code and copy, and ask which tools see your material.
Test with a small project
A short paid engagement shows how the process handles your product, which no demo can.
Our guide to AI design agencies covers the label in more depth, and what a full-service web design agency delivers explains the scope to expect. If you'd rather see our process on your own project, our trial week is one week of real design and development for $1,000.
What hasn't AI changed about good design?
AI hasn't changed what makes a design good: a clear message for a specific buyer, honest proof, a structure that follows the argument and a build that works for everyone. Those were the hard parts before, and they're still the hard parts.
If anything, the bar moved up. When anyone can generate a decent-looking page, decent-looking stops being an advantage. What sets a site apart is the thinking that went into it before the first draft, and the care that went into it after. That's why our startup website design service still starts the same way it always has: with the words. And it's why the work after launch, covered in our guide to the first 90 days after launch, matters as much as the work before it.
