👋 Hello hello,
Every project you re-explained, every preference you re-typed, every "like I said earlier" the AI clearly did not retain. That was you, quietly doing the remembering so the model didn't have to.
This week that started to change, and three companies got there at once. OpenAI, Anthropic, and Perplexity each shipped their own take on giving AI a memory within days of each other, which usually means the whole industry just hit the same wall.
Let's get into what they built.
🔥🔥🔥 Three Exclusive AI Updates
Codex is OpenAI's coding and computer-use agent. Its new Record & Replay feature lets you demonstrate a repetitive task once, like filing an expense report or submitting a time-off request, and Codex turns that demo into a skill.
The useful part is that the skill is inspectable and editable. It saves as a file you can open, tweak, and reuse, so you stop rebuilding the same automation every week. You decide when recording starts and stops.
For now it runs on macOS only, needs Computer Use enabled, and is rolling out to select markets first.
Claude Code is Anthropic's command-line coding agent. The new Artifacts feature publishes your session work as an interactive page at a private link, like a PR walkthrough, a project dashboard, or an investigation timeline.
Here's the clever bit. As the session keeps working, the page refreshes on its own, so everyone you shared it with is always seeing the latest version. It pulls from the full context of your session: codebase, plugins, skills, and connected tools. Pages stay private until you share, and sharing stays inside your organization.
It's in beta on Team and Enterprise plans.
Perplexity Computer is its agent that runs multi-step tasks for you. The new Brain feature builds a context graph of everything Computer does, then reviews it at set intervals to get better over time.
The payoff is that each task starts with context of your projects, decisions, and sources instead of a blank slate. Perplexity says that on tasks needing past context, Brain lifts answer correctness by 25%, recall by 16%, and runs 13% cheaper per task. Every memory links back to the session or source it came from, so you can see and control what it stores.
It's in research preview for Max subscribers, under "Customize" in the sidebar.
🔥🔥 Two Claude Connectors To Test
1. 🎨 OpenArt MCP
One connector that plugs the top image and video models into your agent, including Seedance 2.0, GPT Image 2, and Kling 3 Omni. It works inside Claude, ChatGPT, Cursor, and other MCP clients, and it's the only connector with access to your full OpenArt library. Best for creators who want to generate visuals without leaving the chat.
2. 👀 Visualping
Point it at any web page, tell it what to watch for, and it alerts you the moment something changes, with a plain-English summary of what's different. Alerts come by email, Slack, or SMS. Best for tracking competitor pricing pages, job postings, regulatory filings, or anything you'd otherwise refresh by hand.
🔥 Things You Didn't Know You Could Do With AI
Claude Design is Anthropic's tool that turns a prompt into polished design artifacts. Open Design is the open-source version you run on your own machine, using the coding agents you already have. It's local-first and bring-your-own-key, so your work stays on your laptop.
Download Open Design 0.10.0 from the GitHub releases page (desktop app for macOS or Windows).
Connect a coding agent you already use, like Claude Code, Codex, or Cursor, and add your own API key.
Give it a brief by describing the page, deck, or dashboard you want.
Pick a visual direction, watch it build in the sandboxed preview, and refine it with comments.
Export the finished design to HTML, PDF, PPTX, or MP4.
Don't forget to rate today's post
Until next time,
Team @PracticalyAI
