Every few months there's a new wave of "build a website in minutes with AI" marketing. Wix did it. Squarespace did it. A dozen startups have done it. They all turn out to be fancy templates with a coat of paint. But 2026 genuinely feels different — AI website builders and code-generating models (Claude, GPT-4, Cursor) have moved from impressive demos to tools that actually ship working code. So I ran a real test.

The rules: build a complete portfolio website for a fictional photographer in exactly 2 hours, using AI assistance only. Zero manual coding unless the AI completely failed. From blank page to live URL. Here's what actually happened.

📋
The setupPrimary tool: Claude (best general-purpose coding model in my experience). Backup: GPT-4 for second opinions on tricky CSS. Target: single-page photography portfolio with hero, gallery grid, about section, contact form, dark theme, fully responsive. Hosting: Netlify Drop (free, instant).

Hour 1: The Initial Build

The First Prompt

Opening prompt: "Create a modern photography portfolio website. Single page — hero section with large background image, gallery grid, about section, contact form. Dark theme, minimalist. Responsive. Complete HTML, CSS, and JavaScript in one file."

The result was genuinely surprising. Semantic HTML5, proper heading hierarchy, CSS Grid for the gallery, basic form validation, and alt tags on images — better than plenty of human-built sites I've reviewed. Structure was solid. Accessibility was handled without me asking for it.

The design, though? Looked like a Bootstrap template from 2018. Functional, clean, completely soulless. "Modern" to AI apparently defaults to "inoffensively generic."

The Design Iteration

Follow-up: "Make it visually striking. Add subtle animations, better typography (Inter from Google Fonts), gradient text accents, hover effects on gallery images with smooth transitions."

Noticeably better. CSS transitions appeared. Gradient text worked. Spacing improved. Still had a template quality to it, but now at least a premium template. The trick here: specificity matters enormously. Vague prompts get generic results. "Add hover effects" produces something bland. "Images scale 1.05x on hover with a 0.3-second ease transition and a subtle shadow" produces something usable.

The Content Gap

AI cannot source real photography. I used Unsplash placeholders, which work fine for a prototype but would need replacing for a real client. This is a real limitation that nobody in AI marketing mentions: content is still entirely human work. The AI can wire everything up beautifully. It cannot make creative decisions about visual identity.

Hour 2: Forms, Mobile, and Going Live

The Contact Form Problem

AI wrote a Node.js/Express backend for form handling. Useful in theory. I don't have a server running, and setting one up would blow the 2-hour budget. Pivoted to Formspree — free form backend, works with a single HTML change, 2-minute setup. I asked the AI to adjust the form action and it adapted without complaint. This kind of pragmatic pivoting is where AI assistance genuinely shines.

Mobile Responsiveness

AI wrote all the media queries. But it missed the iPhone SE breakpoint — the navigation collapsed into an unclickable mess at 375px width. One prompt to describe the issue, one fix, working everywhere. The AI wrote a hamburger menu toggle but forgot to write the corresponding JavaScript initially. Caught in testing, fixed immediately. Important lesson: AI-generated code still needs testing. It's not infallible, just fast.

Deployment

Netlify Drop - drag the folder in, live URL in 90 seconds. Free. No account required. A complete stranger could look at a professional-appearing portfolio site from a blank-page start, total elapsed time 2 hours. That's genuinely new.

Honest Assessment: What Worked vs. What Didn't

✓ What Actually Worked

  • Zero to live site in 2 hours
  • Clean, semantic, accessible HTML
  • CSS Grid / Flexbox layouts
  • Responsive foundations
  • Quick pivots (Formspree workaround)
  • Code commenting and organization

✗ What Needed Heavy Assistance

  • Visual personality — all style defaults to generic
  • Complex JavaScript (forgot hamburger toggle)
  • Mobile edge cases (iPhone SE layout)
  • Copy writing — every suggestion was cliché
  • Any actual creative decisions

The copy thing is worth dwelling on. AI suggested "Capturing moments that last forever" for the hero tagline. Every photographer's website says something like that. It took me longer to write one paragraph of actual copy that sounded like a real person than it did for AI to write the entire codebase. Structure is automatable. Voice is not.

🎨
The "vibe" problemThe site worked. It loaded fast. It looked professional. But it had no soul — no quirky detail, no unexpected color choice, nothing that made it memorable. AI optimizes for functional correctness. Delightful surprise is not in the training objective. That gap is significant for anything brand-forward.

AI Tools for Web Development: Compared

ToolBest ForFlexibilityCost
ClaudeCode generation, complex structures, iterationVery high — you own everythingFree / Pro
GPT-4Second opinions, troubleshootingVery highFree / Plus
CursorOngoing development in an editorVery highFree / Pro
Framer AIQuick visual sites for beginnersLow — platform-lockedFree / Paid
Wix AIAbsolute beginners, no codeVery lowFree / Paid

The fundamental divide: code-generating AI (Claude, GPT-4, Cursor) gives you full code ownership, no platform lock-in, and the ability to customize anything. Visual AI builders (Framer, Wix) are faster to start but limit what you can build and charge you forever to keep the site live. For anything you care about long-term, code ownership wins.

Who Is AI Website Building Actually For?

✅ Ready Right Now

  • Personal portfolios
  • MVPs and prototypes
  • Landing pages for validation
  • Internal tools and dashboards
  • Developers who want to move faster

⚠️ Still Needs Heavy Human Work

  • Brand-forward marketing sites
  • E-commerce with complex logic
  • Unique visual identity projects
  • High-stakes client work
  • Sites requiring genuine creative direction

The Workflow That Actually Makes Sense

After this experiment, here's how I'd structure a real project using AI assistance:

Phase 1 AI Structure — ~30 Minutes
  • Generate semantic HTML framework from a detailed prompt
  • Basic responsive CSS with Grid and Flexbox
  • Component architecture and file structure
Phase 2 Human Design — 2–4 Hours
  • Custom visual design decisions — colors, spacing, typography choices
  • Brand imagery and actual photography
  • Micro-interactions and polish that make things memorable
  • All the copy. Seriously, write it yourself.
Phase 3 AI Assistance — Ongoing
  • "Add a dark mode toggle" — AI writes it in seconds
  • "Convert these images to WebP in a shell script" — instant
  • "Write JSON-LD schema markup for these pages" — better than you'd write manually
  • Debugging: describe the problem in plain language, AI finds it

Will AI Replace Web Developers?

After two hours hands-on: no, not in the near term. But the job is changing.

What's automatable: converting a design to code, writing boilerplate components, setting up standard layouts, generating basic form validation. Junior developers whose primary value is "I can translate this Figma design into React" are in a genuinely precarious position.

What is not automatable: understanding what a client actually needs versus what they say they want, making creative decisions with real-world taste, designing systems that scale, debugging non-obvious edge cases in production, anything requiring business context. AI is a multiplier, not a replacer. I built faster. I still made every meaningful decision.

Building something? You'll need tools for your content too.

Compress images before uploading, format JSON APIs, generate QR codes — free browser-based tools, no signup, no limits.

⚙️ Explore Free Tools

Frequently Asked Questions

Yes, to about 80% of professional quality. AI code generators like Claude and GPT-4 produce clean, semantic HTML, responsive CSS, and functional JavaScript quickly. The remaining 20% — brand personality, unique visual identity, polished copy, complex interactions — still requires meaningful human input and judgment.

Conclusion: AI Website Building Is Ready — for the Right Projects

Two hours with an AI website builder produced 80% of a professional portfolio site. That's the honest number. The remaining 20% — the soul, the copy, the visual decisions that make something memorable versus merely functional — was entirely human work.

For side projects, quick prototypes, MVPs, and personal sites, AI is ready right now. Stop waiting and start building. For brand-forward client work or anything requiring unique creative identity, the sweet spot is human-plus-AI: you make the decisions, AI executes them faster than you'd write the code yourself.

The future isn't AI replacing web developers. It's developers shipping in hours what used to take days. That's a significant shift, and learning to work well with AI tools is genuinely the most valuable skill you can build in 2026. If you're managing the files, images, and documents for your web projects, our free browser-based tools — image compressor, JSON formatter, PDF tools — are there when you need them.

Share this review