Our AI process
AI does the legwork. People make the calls.
We use AI agents across research, design, code and QA, inside a design system and rules we write down. Senior people decide what is good, and nothing ships until someone approves it.
What we believe
Six things we believe about AI and design.
- 01
Senior people make the calls. AI does the legwork: research, first drafts, variations, code and checks. Taste, trade-offs and the final yes stay with the senior designers and developers on your project.
- 02
Taste matters more now, not less. AI made average work faster, not better. Anyone can generate a landing page in minutes, and most of it looks the same. Knowing what to keep and what to cut is the job.
- 03
Explore wide, then commit. Options are cheap now, so we look at more of them before anyone commits. Then we pick one direction and finish it properly.
- 04
Systems over heroics. Agents work inside a design system, written rules and checklists, not on vibes. Anything that has to be exact runs as tested code, not a prompt. Small error rates add up: 90% right per step is 59% right after five steps.
- 05
Nothing ships without a yes. Every change comes with a preview. A person reviews it and approves it, and only then does it go live. That goes for our own work and for the agents we set up for you.
- 06
We use it on ourselves first. Our own website runs this way: we ask Claude for a page or a fix, review the change, and ship it. We set up what we use ourselves.
How a project runs now
Six steps from brief to a site your team can run.
Every project is different, but the rhythm is the same. In each step AI takes the repetitive work, and people own the decisions.
01Week 1
Map the real problem
We start with the problem, not just the requested deliverable: goals, users, competitors, your current site or product, constraints, and what needs to move first.
- AI does
- Pulls the brief, call notes, your current site and competitor patterns into one first read, and drafts the audit.
- People do
- Decide what the project is really for, what to leave out, and the order we tackle it in.
- You get
- Project direction
- Website or UX audit
- Sprint plan
- Success criteria
02Weeks 2-4
Explore wide, then pick one
Options are cheap now, so we look at more of them: flows, layouts, copy angles and visual directions, side by side, before anyone commits.
- AI does
- Drafts variations of flows, wireframes, layouts and copy, so we can compare more directions in the same week.
- People do
- Narrow it down to one direction and design it properly. UX runs about a week ahead of UI, so the next flow is ready when the last one is approved.
- You get
- Creative direction
- Wireframes
- High-fidelity design
- Responsive layouts
03Alongside design
Turn the design into a system
Tokens, components and section rules, in Figma and in code. People and agents build from the same parts, so new pages look like your site, not like a template.
- AI does
- Checks screens against the tokens and flags contrast, spacing and naming drift as we go.
- People do
- Set the rules: which components exist, how they combine, and what is never allowed.
- You get
- Design tokens
- Component library
- Section rules
- Brand and tone guide
04Weeks 4-8
Build in small, reviewed changes
Approved design goes to development straight away, so design stays ahead of the build instead of holding it up. Websites in Astro, Framer or Webflow; products in React and Next.js.
- AI does
- Claude Code drafts components and pages from the design system and runs the build. Each change arrives as a pull request with a preview link.
- People do
- Senior developers review every change, fix what the agent gets wrong, and own the architecture.
- You get
- Production build
- Component system
- CMS setup
- A preview for every change
05Week 8+
Review, polish, ship
The last 10% is where the work starts to feel expensive. We test, tighten and polish the details before launch, then stay through it.
- AI does
- Runs the checks on every page: accessibility, contrast, broken links, redirects, structured data and performance.
- People do
- Design QA, responsive checks, copy tweaks and animation tuning, then the final approval before anything goes live.
- You get
- QA passes
- Launch
- Redirect map
- Post-launch fixes
06After launch
Hand over a site your team can run
You get docs for your team and for your agents. If you want it, we set up an AI workflow so your team can ask for pages, updates and reports in Slack or Claude.
- AI does
- Drafts the changes your team asks for, on your design system and content, with a preview.
- People do
- Your team reviews and approves. We stay on for new workflows or a design subscription when you need us.
- You get
- Team and agent docs
- Workflow playbooks
- Training
- Next sprint plan
Timelines are typical, not fixed. A focused landing page can move in weeks; a product redesign, mobile app or full build can take a few months.
The system behind it
We run DesignMe on the same idea.
The company itself runs on a small operating system: written knowledge, AI agents that use it, and code for the parts that have to be exact. It is the same thinking we bring to your design system.
- 01
Written knowledge
Brand and tone guides, SOPs, playbooks and project facts in plain files, which the agents read before they start.
- 02
Skills and agents
Repeatable jobs packaged as skills: drafting, audits, QA and reports. The agent picks the right one for the request.
- 03
Scripts for exact work
API calls, data processing and file changes run as tested code, so they come out the same every time.
- 04
Guardrails and approval
Permissions by role, previews and logs. Agents can draft; people approve what goes live.
- 05
A loop that learns
When something breaks, we fix the tool and update the SOP, so the same mistake doesn’t come back.
What clients notice
Faster first drafts. The same senior finish.
Speed only counts if the work holds up. Timelines and results from our case studies:
4 weeks
To launch Spatial’s new website
3 months
From an idea to N3XT’s MVP launch
300+
N3XT web and mobile screens in about four months
+11%
Conversion across Seamless.AI’s whole site
Keep exploring
- AI workflowsAn agent in Slack or Claude that drafts pages, content and reports. Your team approves.
- Astro developmentA coded site in a repo you own, which your team and agents can change safely.
- AI MVP cleanupBuilt something fast with AI? We give it the structure, states and UX it skipped.
- PricingWhat projects, subscriptions and AI builds cost, with ranges and timelines.
FAQ
Questions about how we use AI
Want to see it on your project? Bring the brief, and we’ll walk you through the first read.
Is your work AI-generated?
No. AI drafts options, code and checks. Senior designers and developers choose, refine and finish what you see, and nothing ships until a person has approved it.
Which AI tools do you use?
Mostly Claude and Claude Code, for research, writing, code and QA. Cursor and v0 for quick prototypes, Midjourney for early visual exploration, and Figma for design. Tools change; the review step doesn’t.
What happens when the AI gets something wrong?
It happens, which is why every change is reviewed. Drafts come with a preview, people check them against the design system and the brief, and only approved work ships.
How often will we hear from you?
Several times a week. We work async in a shared Slack channel, with Loom walkthroughs and focused review calls. You won’t wait two weeks for a big reveal.
How long does a project take?
A focused landing page can move in weeks. A product redesign, mobile app or full build may take a few months. Spatial’s new website launched in 4 weeks; N3XT went from an idea to its MVP launch in 3 months.
Can our team keep using AI on the site after launch?
Yes. We hand over docs for your team and your agents, and can set up an AI workflow so people ask for changes in Slack or Claude and approve a preview before it goes live.
We already built something with AI. Can you work with it?
Yes. We review what exists, keep what works, and clean up the UX, UI, states, frontend and backend logic that fast builds tend to skip. That’s our AI MVP cleanup service.
You've reached the end, so now…
Tell us what you’re building.
We’ll help you map the right scope, team, timeline, and next steps.
Book a call