What an AI-leveraged studio actually does for a client

“We use AI” is the new “we are digital first.” It means everything and nothing.

I want to be more concrete. Here is a walkthrough of what an AI-leveraged studio actually does on a real project, from the first discovery call to the day we ship. Not the marketing version. The actual version.

The example is a recent five-week project: a marketing site rebuild for a B2B SaaS, including positioning work, new copy, design, build, deploy, and a small content launch.

Week 1: discovery and brief#

What the AI does:

• Transcribes the discovery call (Whisper or otterai)
• Pulls out themes, pain points, jobs-to-be-done from the transcript
• Drafts a project brief based on the transcript and our intake form
• Researches the client’s current site, competitors, and category positioning. Generates a competitive landscape document

What I do:

• Run the call. Listen, push back, draw out the real story.
• Read the AI-generated brief, rewrite the parts that are wrong or weak, add what is missing
• Make the strategic calls: who is this for, what is the central promise, what are we explicitly not doing

The AI saves me 4 to 6 hours of post-call admin and research. It does not save me the actual thinking, which is where the value lives.

Week 2: positioning, copy, IA#

What the AI does:

• Drafts the homepage hero variants (3 to 5 versions) based on the positioning
• Drafts every page on the site as a first pass
• Generates the information architecture proposal
• Drafts metadata, page titles, OG copy
• Creates content briefs for the launch blog posts

What I do:

• Pick which positioning angle to commit to. AI is bad at conviction; conviction is what positioning needs
• Rewrite every single page of AI-drafted copy. The first draft is usually 60% there. Editing it down to 90% is the work
• Make the IA call: which pages exist, what they each promise, how they connect
• Apply the brand voice. Generic AI copy has a tell. Removing the tell is the editor’s job

Volume of copy that would have taken two weeks now takes three days. But “three days” still includes my brain, every page.

Week 3: design#

What the AI does:

• Generates initial layout concepts (v0.dev, Bolt, Figma AI)
• Suggests typography pairings based on the positioning
• Creates illustration assets, icons, decorative graphics
• Generates initial image options for the hero and sections

What I do:

• Choose the design direction. AI design is competent but homogenous. Picking something distinctive requires taste.
• Do the actual design work in Figma. AI tools get you to 70% on layout. The last 30% (rhythm, hierarchy, polish) is still craft.
• Commission or curate the photography. Generated images are obvious if you use them poorly. I tend to use them as placeholders, not final assets.

Design is the area where the AI saves the least time. The 70-percent mark is fast. The polish is not. That is fine.

Week 4: build#

What the AI does:

• Scaffolds the codebase (Astro, Tailwind, common components)
• Writes most of the component code in first-draft form
• Generates the Tailwind configuration to match the design tokens
• Catches accessibility issues and CWV problems automatically
• Drafts the deployment config

What I do:

• Architect the codebase. AI is bad at architecture. The choice between content collections vs flat files, where reusable patterns live, what is a component vs a layout. That is judgement work.
• Hand-write the bits that matter: hero animations, complex layouts, edge cases. AI-generated code is usually 80% correct and 100% present. Editing it is faster than writing from scratch.
• Test on real devices. Real Safari. Real Firefox. Real older Android phones. AI cannot test, only generate.
• Performance optimisation. Bundle analysis. Image strategy. Caching strategy.

The build phase is where AI saves the most time. A site that would have taken two weeks of dev work takes one.

Week 5: launch and content#

What the AI does:

• Drafts the launch blog posts
• Generates social copy variants
• Creates the launch email
• Drafts the LinkedIn post for the founder
• Builds the GA4 dashboard with the right events

What I do:

• Edit and rewrite every piece of content
• Time the launch
• Brief the client’s team
• Ship

What stays human#

A pattern across every week:

• Strategy and positioning: AI is bad at conviction
• Taste and judgement: AI is mid-quality on average
• Architecture decisions: AI optimises locally, not globally
• Client relationship: AI cannot read a room

What AI replaces is the production layer. The repetitive, scaling, structured work that used to need junior staff or contractors. That layer is now me with a model. The strategy, taste, and craft layers stay human, because nobody has built an AI that does them well enough to ship.

The economics#

This is why a one-person studio can take on five-week SaaS rebuilds that used to need a team of five. Not because the work is worse, but because the production layer cost effectively went to zero.

The savings get passed on. Or they get reinvested into more attention to the strategy, taste, and craft layers, which is where the work either wins or loses.

That is what “AI-leveraged” actually means at Zivaro. Not “we let AI do it.” More like: “AI handles the slog, so I can focus on the parts of the project that actually need a person.”