Good Morning! Here's what I have for you in today's newsletter:
Anthropic gives Claude one shared memory across chat and Cowork
ChatGPT Work can now log into websites, handling browser tasks for you
Higgsfield launches Ad Multiplier, turning one ad into endless versions
Learn how Codex can control your computer to run real business tasks
4 new AI tools worth trying today
AI TOOLS
Anthropic gave Claude one shared memory across chat and Cowork, letting a Cowork task start from what Claude already knows through prior chats, like a project already discussed, a manager's preferences, or a client mentioned last quarter, with full control over what stays in memory.
Cowork tasks now inherit context automatically from earlier chat conversations, removing the need to re-explain background information every time a new task starts.
Users retain full control over what's actually stored in memory, rather than having every past conversation automatically feed into future tasks by default.
This closes a gap between Claude's two main surfaces, which previously operated with separate, disconnected context despite being the same underlying assistant.
Repeating the same background information every session has been a real point of friction, since chat and Cowork previously operated as if they'd never met. A shared memory removes that friction directly, letting a task start already informed instead of from a blank slate every time.
AI TOOLS
OpenAI rolled out the ability for ChatGPT Work to log into websites on a user's behalf, now live on web and mobile for Plus, Pro, and Business users, extending Work to any task that would normally require opening a browser and clicking through manually.
Work can now handle the full scope of a browser-based task, including the login step itself, rather than stopping short and requiring a person to sign in before it can continue.
The rollout covers both web and mobile simultaneously, meaning the capability isn't limited to a desktop-only workflow.
This is scoped to Plus, Pro, and Business plans, positioning it as a broadly available capability rather than an enterprise-only feature.
A huge share of routine digital tasks still require manually logging into some website to actually get anything done, and that login step has been the wall stopping agents from finishing the job end to end. Removing it means Work can now complete tasks that previously needed a person to at least kick off the browser session.
AI TOOLS
Higgsfield introduced Ad Multiplier, letting anyone take their top-performing creative from Meta and generate endless new versions with different characters, locations, outfits, text, and dubbing.
Any single element in a winning ad, a character, location, outfit, text, or dubbing, can be swapped independently to create a fresh variant.
The tool is built specifically around scaling proven creative, starting from an ad that's already performing rather than generating something new from scratch.
New versions can be used to reach different audiences or extend the life of a campaign that's already working.
Winning ads usually burn out fast once an audience has seen them too many times, forcing teams to start the testing process over from scratch. Multiplying variations from one proven creative instead keeps campaigns running longer and lets a single winning idea reach new audiences without losing what made it work in the first place.
HOW TO AI
Codex is OpenAI's answer to Claude Code, built around a different bet. It lives on your actual computer, not the cloud, controls your mouse and keyboard directly, reads your email and Notion, and turns any workflow into a skill you trigger with one slash command.
Step 1: Install it and pick your setup
Download the Codex desktop app, sign in with ChatGPT. Reasoning level ranges from light to extra high, higher takes longer but answers better. Default model stays in medium reasoning for most work.
Step 2: Give it your computer, browser, and tools
Install Computer Use from the Plugins tab, letting Codex control your mouse and keyboard directly. Add the Chrome plugin for browser control, then connect Gmail, Notion, and Slack the same way, each just a sign-in away.
Step 3: Build a real project and turn it into a skill
Create a project, a real folder everything lives in permanently. Build an agents.md file with real project conventions, this loads automatically every future session:
# Business Expense Dashboard
## Project goal
Turn a folder of real business receipt photos into a categorized Excel dashboard, ready to reference before a quarterly review or tax season.
## Output requirements
One Excel workbook. A raw transactions tab, a summary tab, a dashboard tab with at least two charts.With conventions in memory, the build prompt stays simple:
Read agents.md, then build the expense dashboard exactly as specified
Once happy with the output, reverse-engineer it into a skill:
I'm happy with this output. Turn this into a skill I can run again every time I have new receipts to process.Step 4: Test it and automate it
Have Codex QA its own work using browser control:
@browser use open this dashboard and click through every filter and category, and tell me anything that looks brokenOnce reliable, automate the refresh so new files trigger the skill automatically.
Step 5: Ship it live, for free
Ask Codex to set up a GitHub repository and push the project to it, then connect a free Vercel account to the repo and deploy. Every future push redeploys automatically.
P.S. You can access all the AI trainings, prompts and workflows if you upgrade.
ElevenLabs introduced Composer, a section-by-section song editor that lets anyone start from their own lyrics or track, a blank page, or a prompt, then shape each part of the song piece by piece until it's finished.
Perplexity launched Portable Computer, a fully local version of Perplexity Computer running entirely on NVIDIA DGX Spark hardware with no cloud dependency, powered by a locally installed PPLX 27B model with Qwen 3.8 27B also available.
Vercel launched Vercel Connect, now generally available, giving apps and agents secure, short-lived, scoped access to Slack, Linear, GitHub, and over 100 other services.
🔗 Droid: now connects to Google, Linear, Databricks, Notion, Snowflake, Figma, Datadog, and 50+ more tools through new Connectors and Personas.
🎬 OpenArt: Dub Video, upload one video and get a version for each language, lips naturally synced to every translation.
🦾 Skild AI: S1 now deployed with limited industrial partners, its first step toward general intelligence grounded in the physical world.
🎥 Higgsfield: now in Blender, prompt a scene, describe a camera move, and reblock the whole shot in seconds, available via Higgsfield MCP or Supercomputer.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial