🧠 AI Learns to Watch, Think, and Write Like a Team

🔥🔥🔥 Three big updates

1. Moonshot AI’s Kimi turns a screen recording into a working website

Here’s the entire workflow: record your screen while scrolling through a website. Then type one line to Kimi: Clone this website with all the UX designs.

That’s it.

Kimi K2.5 recreates the layout, the animations, and the small interaction details that usually get lost between design files and dev handoffs.

What’s interesting here isn’t the cloning itself. It’s the input shift. A screen recording suddenly carries the same weight as a full spec document. If this trend continues, prototypes, redesigns, and teardown projects will become much lighter.

Try the demo here.

2. Gemini starts thinking with images, not just about them

Google introduced something called Agentic Vision in Gemini 3 Flash, and the name actually fits.

Instead of looking at an image and answering in one pass, the model now works through it step by step. It plans how to approach the image, runs code to manipulate or inspect it, then looks again before responding.

That loop makes image understanding feel less like pattern matching and more like investigation. Google reports a steady quality bump across vision benchmarks, but the bigger shift is behavioral.

This opens the door to more reliable visual work: design checks, technical diagrams, dense charts, anything where one glance isn’t enough. If you want to know more about how this works, read their blog here.

3. Prism gives scientists a real AI writing space

OpenAI launched Prism, a free workspace built for scientific writing and collaboration, powered by GPT-5.2.

Instead of bolting AI onto existing tools, Prism starts where researchers already spend their time. Writing papers. Sharing drafts. Working with collaborators. Anyone with a ChatGPT account can use it. Unlimited projects. Unlimited collaborators.

Try it for yourself here.

🔥🔥 Two Tools Worth Trying

1. 🧪 Kilo Code Reviewer

Open a pull request, and Kilo jumps in immediately. It runs an AI-powered review before anyone on your team even looks at the code.

This works well for teams that move fast and don’t want small issues piling up. It’s not trying to replace human review, just making sure the obvious stuff doesn’t slip through.

2. ⚠️ Skills.sh

Skills.sh is an open agent skills ecosystem comprising a free collection of reusable skills for AI agents. Think of it as a toolbox you can pull from instead of rebuilding the same capabilities over and over. It’s great for experimenting and learning how agents behave in the wild.

But just be careful. Shared skills also mean shared risk, especially around prompt injection.

Explore it here.

🔥🔥 Things You Didn’t Know You Can Do With AI

Find leads, redesign their site, and send the outreach — in one flow

Cold outreach usually starts with an email.

This workflow starts with actual work.

Using Claude Cowork, you scan for businesses that already have strong reviews but weak websites. The system audits their site, flags issues, then designs and codes a new version that fixes them.

That redesigned site gets deployed on Netlify. Then an outreach email is drafted, including the audit and the live redesign.

Here’s the prompt used in this workflow.