Good Morning! Here's what I have for you in today's newsletter:
Anthropic merges Claude Cowork and chat into one Claude
xAI gives Grok Build cross-session memory for conventions and project facts
Google brings full Gemma 4 12B support to 16GB Macs
Learn the five free repos that give Claude Code actual superpowers
4 new AI tools worth trying today
AI TOOLS
Anthropic merged Claude Cowork and chat into a single Claude experience, rolling out to Pro and Max plans on web, desktop, and mobile over the coming weeks, with more plans to follow.
Any conversation can hand off a long-running task such as a report, spreadsheet, or presentation, and Claude keeps working after the laptop closes, asking for input when something is unclear.
Claude Docs and Claude Slides launch in beta alongside the merge, letting users draft, comment on, and edit documents and decks inside the conversation, then export to Word, PowerPoint, or PDF.
Claude Design runs inside conversations as well, and existing Cowork chats, projects, artifacts, connectors, and skills carry over automatically with nothing to switch on.
There is no mode to pick before starting a task. Ask the question, or hand over the deck due at noon, and Claude picks the tools, runs it in the background, and delivers a file you can edit in place. If your message box still shows Chat and Cowork, you don't have it yet.
AI AGENTS
xAI added cross-session memory to Grok Build, its terminal coding agent, so conventions, decisions, and project facts are recorded as markdown notes and read back in later sessions.
Notes are written in the background while work happens, split into global files like MEMORY.md and preferences.md and workspace files such as testing.md, code-style.md, and gateway.md.
When a user corrects the agent, for example that tests must run through a specific command that starts the database first, the convention is saved and applied before touching related code.
The /memory command opens a browser to view, search, and delete saved notes, /dream consolidates recent notes into topic files, and /remember saves a note directly.
A coding agent that forgets your project when the terminal closes makes you re-explain the same commands and rules each morning. Grok Build keeps that in plain markdown you can read and edit, so the second week on a codebase costs less than the first.
AI MODELS
Google updated the AI Edge Gallery app for Mac with full Gemma 4 12B support, bringing its multimodal open model to Apple laptops with 16GB of memory, including the MacBook Air.
The model accepts vision and audio input alongside text, with configurable vision token budgets so users can trade image detail for speed on tighter hardware.
Multi-Token Prediction generates several candidate tokens per pass to cut decode latency, and a 16GB M4 MacBook Air produces about 15 tokens per second in testing.
Gemma 4 12B is Apache 2.0 licensed, delivers performance close to Google's 26B mixture-of-experts model at under half the memory footprint, and runs Python locally to return charts inside the chat.
A base MacBook Air can run a 12B model that sees, listens, and writes code with the Wi-Fi off. Speed is modest, around 15 tokens a second, but everything stays on the machine, so sensitive documents, meeting audio, or client screenshots never leave your laptop. The download is free.
HOW TO AI
A harness is everything you bolt onto Claude Code before it ever touches your task, the standing rules, the checks, and the files it reads first. Four of these repos combine for over 700,000 stars, and the fifth is the answer for anyone who wants control instead of a full system taking over.
Repo 1: Superpowers
The most-starred and most opinionated. The moment it sees you building something, it stops, works a spec out of the conversation, shows it to you, and only writes an implementation plan once you sign off.
/plugin install superpowers@claude-plugins-officialThis can burn through tokens fast, it owns the whole loop from idea to reviewed code.
Repo 2: The Karpathy guidelines
A community file built from Andrej Karpathy's own post listing how coding agents tend to fail. Four rules: think before coding, keep it simple, make surgical changes, stay focused on the goal.
curl -o CLAUDE.md https://raw.githubusercontent.com/multica-ai/andrej-karpathy-skills/main/CLAUDE.mdThe lowest-commitment option here, merge it into your existing CLAUDE.md rather than replacing it.
Repo 3: GStack
Garry Tan's own Claude Code setup. Splits work into roles, product vision, designer, engineering manager, release manager, doc engineer, QA, each behind its own slash command.
git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack
cd ~/.claude/skills/gstack && ./setup
Repo 4: Ponytail
Forces the smallest solution that works, YAGNI, standard library first, no abstractions nobody asked for. Benchmarked against a real repo: roughly 54% less code on average across twelve tickets, up to 94% where an agent would've over-built something.
/plugin marketplace add DietrichGebert/ponytail
/plugin install ponytail@ponytailSend these as two separate prompts, sending them together can fail the install silently.
Repo 5: Matt Pocock's skills
The deliberate opposite of Superpowers. Small, composable skills instead of one whole process.
npx skills@latest add mattpocock/skillsHis most popular one, grill-me, interviews you about what you're building before anything gets written. caveman cuts token costs. Install only the ones you want.
P.S. You can access all the AI courses and trainings (including the full version of this one), prompts and workflows if you upgrade.
Meta released Muse Code natively on Windows with no WSL required, running PowerShell-fluent, sandboxed by default, native on x64 and ARM64, and installable without admin rights via a single PowerShell command, with intersession messaging coming next.
Cognition introduced Code Scans for Devin, codebase-wide audits for any goal such as finding dead code or slow queries, where Devin investigates across the repository, reports findings, and opens the pull requests, powered by Agentic MapReduce.
ElevenLabs launched Reception, an AI receptionist platform for small businesses built on ElevenAgents that answers every call, handles questions, books the job, and texts a confirmation, with setup taking minutes after adding a website URL.
🎞️ Motion: 13.4 adds AnimateView, a view transitions component with spring support, dynamic targets, and animation controls.
🔀 OpenRouter: added Union Alpha, a free stealth model with 256K context, multimodal input, and the provider doesn’t train on prompts.
⌘ Command Code: launched a desktop app with multiple agents, Git, terminal, browser previews, and instant edits via /design.
🎨 Canva: connects to Claude for Small Business, turning workflows into on-brand assets.
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