For writers and creatives, maintaining a polished portfolio is a constant balancing act. Updating static PDFs requires design skills, while generic platforms often lack personality. Enter Claude Design, a new feature from Anthropic that promises to generate visual assets, including portfolios, directly from text prompts.
I tested this capability to see if I could replace my outdated website and rigid PDFs with a custom, stylish portfolio. The result was a stunning first draft that arrived in minutes—but it came with a critical warning: AI is excellent at aesthetics, but dangerous with facts.
The Setup: Context Is King
To begin, I subscribed to Anthropic’s Pro plan ($20/month) to access Claude Design. While the desktop app offered more features, I used the browser version due to storage constraints. The interface is straightforward: a chat window on the left and a design canvas on the right.
The key to success lies in context. Claude Design performs best when provided with specific materials. I uploaded:
* A professionally designed PDF of my 2020 portfolio (for style reference).
* A Word document containing links to my recent articles.
* Specific images for key stories.
My prompt was precise: “Create a new editorial portfolio to send to prospects. Maintain the look/feel of the attached PDF but add all new clips linked in the Word document. Use New York Mag (nymag.com) as design inspiration.”
I selected the latest model, Opus 4.7, for maximum capability.
The First Draft: Style Over Substance
Within five minutes, Claude generated a portfolio that was visually striking. It successfully mimicked the clean, editorial aesthetic of New York Mag, proving that AI can indeed replicate high-end design sensibilities without manual coding or graphic design software.
However, the content was riddled with hallucinations.
* False Biographical Data: The AI claimed I was from Sydney, a city I have never lived in.
* Fabricated Work History: It listed publications I had never written for.
* Invented Reception: Perhaps most dangerously, it made up details about how my articles were received, creating fictional praise and metrics.
Key Takeaway: AI design tools are generative, not factual. They prioritize visual coherence over truth. If you do not verify every single line of text, you risk presenting a beautiful lie.
The Correction Process: Iterative Refinement
The true test of any tool is how easy it is to fix its mistakes. Claude Design offers two methods for revision: Tweaks (via chat) and Edit (direct manipulation).
- Quick Fixes: I used the “Tweak” feature to correct my location. The change was instant and seamless.
- Image Integration: The AI initially failed to pull images from my linked articles. However, when I explicitly uploaded a photo and instructed it to use that specific asset for a story, it complied perfectly. This suggests that direct file uploads are more reliable than link scraping for visual accuracy.
- Manual Editing: Using the “Edit” function, I opened the live file to correct my bio. I stripped away the fluff and focused strictly on my journalism credentials.
The editing process was intuitive and did not feel like “fighting the tool,” a common complaint with other AI platforms. The interface allowed for granular control, letting me adjust copy and layout without starting over.
The Verdict: A Powerful Assistant, Not an Autopilot
Claude Design is arguably the most impressive AI creative tool I have used. For $20 a month, it delivered a professional-grade prototype that was 75% complete on the first try.
Why this matters:
* Speed: What might take a designer days can be prototyped in minutes.
* Accessibility: Users without design skills can achieve high-end visuals.
* Risk: The potential for factual errors is high. The tool is a co-pilot, not an autopilot.
I will finish the manual fact-checking and use this portfolio for future job applications. The next step? Asking Claude to build a full website to host it. But until then, the lesson is clear: Let AI handle the style, but keep a human hand on the truth.
