Good Morning! Here's what I have for you in today's newsletter:
ChatGPT Work now learns your writing style from your own emails and files
OpenDesign's new Harness hits state-of-the-art quality
Supermemory introduces learner-1, pushing the frontier for agent memory
Turn Claude Code into a tutor that can teach you anything
4 new AI tools worth trying today
AI TOOLS
OpenAI added the ability for ChatGPT Work to learn a person's actual writing style, picking up favorite phrases, specific sign-offs, and capitalization habits by connecting tools like Gmail, Google Drive, Slack, and SharePoint.
ChatGPT Work reads across emails, messages, and files from connected tools to build a picture of how someone actually writes, rather than relying on a single sample.
Once learned, the style carries into whatever gets written next, applying the same phrasing patterns and habits automatically.
Connecting multiple tools at once, rather than a single source, means the model draws from a broader range of someone's actual writing.
ChatGPT Work pulls directly from real emails and messages instead of a self-described style, capturing actual writing patterns as they exist. This removes the step of manually specifying tone or formatting preferences, since the model already has that information from real usage.
AI AGENTS
OpenDesign introduced its new OpenDesign Harness, achieving state-of-the-art design quality at roughly a twentieth of the cost of its prior approach, with over 88% of survey respondents rating the results positively, on by default for everyone starting tomorrow.
The harness can turn a static presentation into an interactive demo that feels like the actual brand, letting visitors explore details at their own pace instead of scrolling through fixed slides.
Designs follow a brand's own brief and design system, with imagery matched to the specific product and business rather than a generic layout template.
It builds complete, clickable apps with real navigation, multi-step flows, and state that carries across pages, going beyond a single first-screen mockup.
A 20x cost reduction alongside a quality improvement is a rare combination, since efficiency gains typically come with a tradeoff in output quality. Building a full clickable prototype with real navigation used to mean spending days in a tool like Figma. Someone can now hand this a brief and get something they can click through and test with actual users, without building it by hand first.
AI RESEARCH
Supermemory introduced learner-1, its push into continual learning and in-context memory for agents, built on a system that already claims sub-300ms recall, deduplicated token billing, and top scores on benchmarks like LongMemEval and LoCoMo.
Supermemory's existing memory graph merges, updates, and contradicts facts across sessions instead of treating each new chunk as a permanent, isolated entry the way a standard vector database does.
Re-storing the same content, a synced document or a repeated conversation, costs nothing extra, since only genuinely new tokens get billed.
The company reports recall times up to 10x faster than Zep and 25x faster than Mem0 in its own published benchmarks.
Deduplicated billing removes the cost penalty for an agent that loops over the same context repeatedly, checking a synced doc or replaying part of a conversation. A memory system built to merge and contradict facts over time catches outdated information directly, instead of surfacing it alongside whatever's current.
HOW TO AI
Asking Claude a question gets you an answer, not a course. A skill called /teach, built by Matt Pocock, fixes that. It turns a folder into a standing course on one topic, sourced first, not taught from memory.
Step 1: Install the skill
npx skills@latest add mattpocock/skills --skill=teachThen inside Claude Code: /teach. Use a dedicated repo, not a catch-all folder.
Step 2: Give it a mission
It interviews you first, saved as MISSION.md.
I want to learn conversational Japanese well enough to hold a real conversation with my partner's family within six monthsStep 3: Confirm it's citing sources
Show me RESOURCES.md
A lesson with no citations is the skill teaching from memory, exactly what it's supposed to avoid.
Step 4: Take your first lesson
/teach lesson oneOne sitting, one win.
Step 5: Come back for lesson two
Close the session, open a fresh one in the same folder.
/teach next lessonThe folder remembers, not the conversation.
Step 6: Know where it fails
The correct quiz answer lands in the first slot almost every time. Files can also land in the wrong folder. It can fabricate specifics with confidence too.
Open the primary source you cited for this lesson and confirm it actually says what the lesson claims
P.S. You can access all the AI trainings (including the full version of this one), prompts and workflows if you upgrade.
Arm introduced CSS for Mobile 2, an AI-native compute platform built for the system-level demands of agentic AI and cinematic mobile graphics, combining a new AI CPU and GPU to change what the next generation of mobile devices can do.
fal added native, real-time Full HD generation for the H3 Max family, with both H3 Max and H3 Max Turbo now producing 5 seconds of 1080p video in under 5 seconds of generation time.
Cua enabled Cursor Cloud Agent workers to run on Windows and Linux through Cua Cloud Fleets, with Cursor running the agent loop while a Cua VM executes the actual tools.
🎨 Motion: now animates Three.js objects, materials, and shader uniforms sequenced alongside HTML and SVG elements, in Motion+ Early Access.
👆 Rork: Element Selection lets anyone select any element in the preview and tell the agent how to change it, no screenshot needed.
🤖 HeroUI: Agents embed directly into a product, understanding page context, calling tools, and generating charts, tables, and forms.
🖼️ Framer: GPT-6 Astra is now available, designing creative components and elegant layouts with five reasoning levels and a fast mode.
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