Good Morning! Here's what I have for you in today's newsletter:

  • Claude now has its own built-in browser, navigating web tasks in Cowork

  • Higgsfield introduces Relight, adding professional lighting to any shot

  • Google brings Gemini 3.5 Transcribe, its most precise speech-to-text model

  • Build a full copywriting system inside Claude with six real skills

  • 4 new AI tools worth trying today

AI AGENTS

Anthropic gave Claude its own built-in browser inside Cowork, opening directly in the side panel whenever a task involves a website, so Claude can navigate, fill out forms, and finish the job without leaving the app.

  • The browser opens automatically in Cowork's side panel the moment a task requires visiting a website, rather than requiring a separate manual step to launch it.

  • Claude can navigate pages and fill out forms directly inside this browser, completing multi-step web tasks end to end rather than stopping partway through.

  • This keeps the entire task inside Cowork itself, avoiding the need to switch to an external browser tab to finish work Claude started.

A separate browser extension acts inside a sandboxed tab, cut off from your actual session, cookies, logged-in state, autofill. Cowork's native browser inherits that real session context instead, so a task like "renew my domain" or "submit this expense report" runs straight through instead of stalling out on a login screen halfway.

AI TOOLS

Higgsfield launched Relight, letting anyone add professional lighting to any shot after it's already been filmed, dragging lights around a subject from any angle and adjusting color and brightness for each one.

  • Lights can be positioned from any angle around a subject after the fact, rather than requiring the lighting setup to be planned and executed during the actual shoot.

  • Each light source supports independent color and brightness control, giving fine-grained adjustment rather than a single blanket lighting effect applied to the whole frame.

  • The tool can change the mood of an entire scene through lighting alone, without needing to reshoot or re-edit any other part of the footage.

A full AI restyle tool changes everything in a shot at once, color grade, texture, sometimes the subject itself, which makes it unusable when the footage is already good and only the lighting is wrong. Isolating lighting as its own adjustable layer keeps the rest of the shot untouched, closer to how a colorist works than how a generative video model usually operates.

AI MODELS

Google introduced Gemini 3.5 Transcribe, a new speech-to-text model built for precise, intelligent transcription, better at handling complex phone numbers and postal codes in noisy environments, removing filler words automatically, and recognizing custom vocabulary across more than 85 languages.

  • The model specifically improves accuracy on structured spoken data like phone numbers, postal codes, and order IDs, even when background noise would normally interfere with recognition.

  • Custom vocabulary recognition means the model can learn unique names and product titles specific to a business, rather than defaulting to generic transcription that misses unfamiliar terms.

  • Support spans more than 85 languages with automatic detection, removing the need to manually specify which language is being spoken before transcription starts.

Custom vocabulary support is the detail that actually separates this from a general Whisper wrapper, since most transcription APIs charge extra or require fine-tuning to learn a company's own product names and terms. Baking that into Gboard and the Gemini app directly, rather than gating it behind an enterprise API tier, puts a capability normally reserved for call-center vendors into a free consumer keyboard.

HOW TO AI

Copywriting was never really about the words. It's a way of thinking, a process you run before you ever write a sentence, the strategy, the sentence-level rules, the structure an idea actually needs.

This guide turns that process into something Claude actually runs for you, six real skills built from a real, working framework.

Step 1: Build the strategy skill, the 6Ps

Before any copy gets written, six questions get answered: People, Positioning, Promise, Proof, Priority, and Process. Skip this and you get technically fine sentences that convince nobody, since you never figured out who you're talking to.

/plugin marketplace add anthropics/claude-code
/plugin install skill-creator

Build a skill called "copy-strategy" that interviews you through the 6Ps on a real product, then saves it as /copy-strategy.

Step 2: Build the craft skill, the 5Cs

Strategy tells you what to say. The 5Cs, Clear, Concise, Concrete, Conversational, Cadence, make a sentence actually land once you're saying it. Build a skill called "copy-craft" that edits any draft against them, showing the original line next to the edit.

Step 3: Score the hook before you post it

Every real hook does three jobs: tells a stranger what it's about, shows what they'll get, and makes them need to keep reading. Build a skill called "hook-score" that checks the three-job test first, then scores Benefit Γ— Relevance Γ— Credibility Γ· Perceived Effort.

Step 4: Pick the right structure

Different ideas need different shapes. Build a skill called "copy-structure" offering all three: the Micro Story, the Pyramid Principle, and Cross-Domain Synthesis, showing all three versions of the same idea side by side with a recommendation.

Step 5: Build your greatest hits

Every good writer has 8-10 ideas they keep returning to. Build a skill called "greatest-hits" that reviews your past posts and identifies the recurring ideas that performed best, refining each into one sharp sentence.

Step 6: Chain it all into one command

Build a skill called "write-copy" that runs a request through copy-strategy, then copy-structure, then copy-craft, scoring the final hook with hook-score, and saving the finished piece as a real markdown file:

/write-copy: a LinkedIn post about why most agencies undercharge for strategy work

P.S. You can access all the AI trainings (including the full version of this one), prompts and workflows if you upgrade.

Anthropic added the Admin API to its SDKs and the ant CLI, letting admins manage members, workspaces, and API keys, and read an organization's rate limits directly.

Qwen released Qwen3.8-Flash, a multimodal MoE model and early preview of the Qwen4 architecture, trained at a fraction of the cost of its predecessor while outperforming it across coding and office tasks.

Ollama shipped v0.33, letting Claude Desktop connect to Ollama as a third-party gateway provider with one toggle, running cloud and local models side by side with built-in web search.

🎭 Higgsfield: swap yourself into anyone, change backgrounds, or recreate viral effects with just one photo using the Seedance 2.5 model.

πŸ”— Grok: Linear triggers now live, connect Linear to triage issues, track work, and auto-start when an issue is assigned to you.

πŸ’¬ X Chat Agents: a new Chat API and skeleton framework for building a Chat Agent from scratch.

🎨 Antigravity: Interactive Generative UI Artifacts, rich data visualizations and 3D simulations rendered inline.

Which image is real?

Login or Subscribe to participate

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

βœ“ Full archive of premium guides with ready-to-use prompts

βœ“ Structured AI courses (step-by-step, start-to-finish)

βœ“ Every upcoming premium tutorial