It's Monday. Here's what changed in AI this week, and what it means for your design work:

The start of the week was all about the launch of Stage V2. I've been deep in user calls - listening, asking what's broken, and what people want next. The feedback has been genuinely positive, which has me excited to keep building. Right now the focus is the quality of the high-fidelity frames, which honestly isn't where it needs to be yet. Expect a big jump there in the next few days.

On the Limora side, I'm testing a new client-acquisition play built around trial reels - something I've never done before, so I'm still experimenting. The nice part: I can post a lot of reels promoting the tool directly from my own accounts, so people see it's coming from me. They don't show to my followers, so I'm not constantly talking about Limora in your feed - I don't want to bore you with it. Those were the two big pushes this week, on top of content.

For two years, the rule was simple: use Claude for frontend. That rule just cracked.

OpenAI shipped GPT-5.6 on July 9 - a three-model family: Sol (the flagship), Terra (mid-tier), and Luna (fast and cheap). For designers, the headline is the frontend jump. On OpenAI's own seven-task frontend benchmark, GPT-5.6 scored 4.4 on their five-point QA rubric - up from 4.0 for GPT-5.5, and ahead of Claude 4.8's 3.5. With only high-level direction it builds tasteful, functional interfaces, and its stronger computer-use means it inspects the rendered result and fixes visual issues before handing it back - not just generating code and hoping.

The interesting part isn't the benchmark, it's the design judgment. Reviewers say GPT-5.6 now makes tasteful calls on layout, spacing, and hierarchy on its own - the thing GPT always used to get wrong. It's not a clean sweep. Claude Fable 5 still edges it on visual polish and knowing when to hold back - one test had GPT over-animating a blog page where Fable correctly kept it calm. But the gap that used to be a canyon is now a single step. And when two frontier models are this close on design, the winner stops being the model. It's the designer who knows how to direct them.

Don't switch blindly. Run your next UI build through both Sol and Fable 5 with the same prompt and compare the output side by side. Reach for Sol when you want it to self-correct the rendered result; reach for Fable when taste and restraint matter most. Start in ChatGPT with GPT-5.6.

Meta finally shipped an image model worth caring about - and it's already inside the apps you post to.

Muse Image is Meta Superintelligence Labs' first image model, now live in Meta AI, Instagram, and WhatsApp. It reads complex prompts, blends multiple photos into one clean image, and lets you sketch edits directly on the result. It also renders readable in-image text, restores old photos, and can redesign a room using real products you can actually buy.

Turn any website into a fully editable design in about 10 seconds - real HTML and CSS, not a screenshot. Here's the workflow with Paper.

💌 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,

Say hi on X, Instagram or YouTube