It's Monday. Here's what changed in AI this week, and what it means for your design work:
🧪 What I'm Building: A full studio week and a new brand in the AI space
🚨 Big News: Kimi K3, the open model that just topped front-end design
🤖 Prompts Inspiration: Premium campaign design with the exact prompts
⚙️ Tool of the Week: GenSpark Design
🛠️ Tutorial of the Week: Run a team of AI agents at once with Grok Build
🔗 Quick Links: 5 things that caught my eye
WHAT I’M BUILDING
Behind the Scenes
Busy week, and most of it was studio work. We're closing out a project for one of our branding clients right now, the kind of finish where every last detail matters. And we just signed a new one. A really cool company in the AI space that wants a brand identity strong enough to push them to the next level. That's the fun part of this work, taking something with real potential and giving it a look that matches.
The other track, it’s the software. We're still jumping on calls with users every week, just listening. What do they actually want us to build? You learn more from ten of those conversations than from a month of guessing. More on both soon.
Also went on a short weekend in Paris. Such a great city to visit in the summer.
BIG NEWS
Kimi K3, the Open Model That Just Topped Front-End Design
This is the first open model I'd actually trust to design a front-end.
Last week Moonshot AI released Kimi K3, a 2.8 trillion parameter open-weight model with a 1 million token context window. Within hours of launch it jumped from #18 to #1 on the Frontend Code Arena leaderboard, leading in six of the seven front-end domains tracked and overtaking Claude Fable 5, the model most builders treated as untouchable for UI work. It's also natively multimodal, so it reads text, images, and video in the same model.
My take
Front-end is exactly where most AI models fall apart. They generate layouts that look generic or break the second you resize the window. An open model topping the front-end board changes the math for designers and builders. You can point it at a screenshot of a site you like and get clean, responsive HTML and CSS back. The 1M context means it can swallow your entire design system and component library in one prompt, then generate new screens that actually fit your patterns instead of inventing their own. The sleeper feature is motion. Because it understands video natively, it's strong at animation and micro-interactions, not just static screens. Open weights also means the price floor drops and you can run or fine-tune it yourself.
How to use it
Try it on Kimi Code or the Kimi app. Drop in a screenshot of a layout you admire and ask for a responsive front-end, or feed it your existing component library and have it build the next screen in your system. Start there before you reach for a paid model.
PROMPTS INSPIRATION
3 Prompts Worth Stealing This Week
Instagram campaign poster
Act as the Global Creative Director for Parle Hide & Seek, thinking like a collaboration between Parle Creative Studio, The Minimalist, Pentagram, Buck...Minimal Grain Speedline Sports Poster
One design language locked across four sports posters: extreme negative space, high-contrast motion lines, single accent color per study, consistent...Editorial zen environment portrait
High-fashion editorial portrait, zen environment, adult East Asian woman, theme 'Stillness / Zen'. Real photographic subject + minimal zen interior + large negative space...TOOL OF THE WEEK
GenSpark Design
Going from an idea to something you can actually show still eats days. Screens, then a landing page, then handoff. GenSpark Design collapses that into one step.
Type a single sentence, something like "design a note-taking app," and it builds polished, finished app screens, a matching landing page, a product demo video, and then converts the whole thing into working, deployable code. Not a mockup. A running product you can ship or hand to a developer.
How I'd use it
Pitch faster: generate a full app plus landing page for a client idea before the kickoff call, so you walk in with something real.
Launch in one pass: spin up a product page and a demo video together instead of stitching three tools.
Scaffold, then refine: take the deployable code as a starting point and polish the details in Figma or your own stack.
TUTORIAL OF THE WEEK
Run a Team of AI Agents at Once With Grok Build
Most AI builders make you wait while one agent works through your task. Grok Build runs a whole team of agents at once. Here's how to set it up.
Step 1: Install Grok Build
Run
npm install -g grok-buildin your terminalRun
grok-build initto start the local agentIt spins up a CLI with an optional web UI, synced over WebSocket
Step 2: Describe the full task
Open the project you want to work on
Step 3: Run the agents in parallel
Add the
--parallelflag to spawn up to 8 sub-agentsEach one works in its own Git work-tree on a separate branch, so nothing collides
Step 4: Let the orchestrator merge
A coordinator splits the work, the agents run at the same time (plan, search, build)
The orchestrator reconciles their outputs and hands you clean diffs to review
Step 5: Review and ship
QUICK LINKS - This Week's Radar
Thinking Machines 'Inkling' - Mira Murati's startup shipped its first model, a 975B open-weight release (July 16).
OpenAI Codex Micro - OpenAI's first hardware, a $230 physical keypad for driving your Codex agents.
Google Video Remix - Gemini inside Google Photos relights, swaps backgrounds, and restyles 10-second clips.
Magnific Figma plugin - generate, upscale, relight, and remove backgrounds without leaving Figma.
Seedream 5.0 Pro - ByteDance's image model splits a render into editable layers you drop straight into Figma.
💌 Know a designer who should be using AI smarter?
Forward them this email. Or just send them to logiaweb.net to join.
See you next Monday,
