Using Claude, Cursor, and modern AI tools to go from design to shipped product.
26 articles
We gave our 3D blog museum to Kimi K3.1 and Opus 5.5 with the same short prompt. One worked for 8 hours, the other for 70 minutes. Here is what each built, why 1 reference image mattered most, and the prompt to copy.
We asked Claude Code to study Apple's Passwords app with REA before it wrote any code. It found 17 rules with evidence, including 1 we would never have guessed. Here is the research, the steps, and 1 correction to the video.
Our first AI-built Luis Barragán tribute site looked like every other AI website. So we rebuilt it with 7 steps instead of 1 prompt, then checked the research behind each step. Most of it held up. 1 step needed a correction.
We asked Claude to audit our own chats, then built 6 Claude Code mods without writing code. Each one comes with a prompt you can copy and use free.
Google announced Gemini 4 Argon on 30 September 2026. We are designers, so we read the benchmarks the slow way, checked what we said on camera, and turned the announcement into an animated explainer with 1 prompt to Claude Opus 5.5. Here is where Argon leads, where it trails, what a 1% gap means, and how to make your own explainer.
OpenAI launched ChatGPT Dots on 29 September 2026: AI agents that keep working when you close the laptop. We are in the UK, where Dots aren't available yet, so we checked the launch and the research on delegation instead. Here is what Dots are, where they sit next to Codex and Grok Bot, and 6 rules for delegating without getting average work back.
Anthropic shipped Fable 5.1 on 1 September 2026. A creator I follow spent around $700 on the API testing it for one video. I had a laptop, a weak connection and a normal subscription, so I ran 3 open briefs instead. All 3 prompts are here to copy, and all 3 results are live.
There is no separate Grok Bot subscription. As of 26 August 2026 the cheapest route in is Cursor Pro at $20 a month. Here is what that gets you, why you start with 1 agent and end up with 6, and the if-then line that turns a weekly list into work you actually do.
You wrote a CLAUDE.md 6 months ago and added a rule every time an output came back wrong. Nobody has deleted a line since. When Opus 5 shipped, more than 80% of Claude Code's system prompt was deleted and the model got better. Here is the research behind that decision, and the protocol for running the same delete on your own file.
Can Hoskan built a Messi World Cup tribute site in the Claude chat window, then we ran the identical prompt through Claude Code on low and high effort. Low was delete-it-and-start-again bad. High beat everything on the desk, including his chat version.
2 models on 2 screens, 3 briefs, 1 shot each, no corrections allowed. A pricing page, an Apple Weather clone, and an agentic task decided a 2 point gap — and it wasn't speed that made the difference.
Every new chat, Claude forgets who you are. A second brain is one folder it reads first, so every session starts with your context already loaded.
AI agents are already browsing your website. Most sites give them nothing useful. Here's how to fix that.
Claude Design burns tokens fast and locks you into one ecosystem. Open Design is the free, open-source alternative. Here is what it does, how the output compares, and who it is actually for.
How the 02UI Labs team went from a Figma sketch to a live product in one day using Claude Code and Cursor.
The workflow has changed enough that the gap between an idea and a working product can now be measured in hours. Here's what that actually looks like when you run it without a coding background.
Same brief, same conditions, two AI design tools. Here's what we learned about the tool that actually matters most.
Same screenshot, same prompt, two different models. We tested what actually changed for designers using Claude to generate UI.
Only 1% of designers are using AI effectively. Here's why you're still early, and what to actually do about it.
AI coding tools are fast. Without context, they build generic. DESIGN.md fixes that.
Figma published a blog post about AI agents on the canvas. Most setup guides assume a terminal. This walkthrough is for designers who want to test it without a developer setup.
You spend hours polishing spacing, type, and colour in Figma. Then the coded version comes back "close enough." Here’s why that happens, and a workflow that reads real design data before writing code.
Designs that shine visually but frustrate users? You're not alone. Clashing colours, poor hierarchy, ignored accessibility these blind spots plague solo Figma
AI updates are constant. Every week there is a new tool, a new feature, a new debate. If you spend even 20 minutes on social media, you will see questions
Figma just released 3 new AI image editing tools that make it much easier to work with photography directly on the canvas.
As designers, we spend years refining our eye studying color, typography, and layout but real growth often comes from good feedback. The problem is, most of us
