
Good Morning! Here's what I have for you in today's newsletter:
Anthropic connects Claude Managed Agents to Vercel's Chat SDK
Higgsfield brings native video upscaling to 2K and 4K with FLUX.3
Google introduces Gemini Omni 1.1 Flash, newest model for generative video
Turn Claude into your own free website designer with today's guide
4 new AI tools worth trying today
AI AGENTS
Anthropic added a new cookbook connecting Claude Managed Agents to Vercel's Chat SDK, giving an agent access to a universal chat layer with one type-safe handler and 15+ adapters across Slack, Teams, Discord, and web, while Managed Agents handles the server-side harness, session management, and memory.
Chat SDK provides the interface layer, a single onDirectMessage handler that works across every supported platform, rather than requiring a separate integration for each one.
Managed Agents runs the agent logic server-side, keeping the harness, session state, and memory separate from whichever chat surface the message came through.
Each conversation persists as one continuous session, meaning an agent doesn't lose context switching between a Slack thread and a web chat if both route through the same integration.

This is a cookbook, not a shipped product, so it's really a working example showing how to wire a Managed Agent's session state into Vercel's Chat SDK without building that plumbing yourself. Anyone already running both stacks gets a documented starting point instead of reverse-engineering the handshake between agent memory and a chat adapter.
AI MODELS
Google introduced Gemini Omni 1.1 Flash, bringing a new suite of creative controls and generative video capabilities to developers, including scene extension for longer storytelling, first and last frame specification, efficient 360p drafting, upscaling to 4K, and video references in multimodal input, now available via the Gemini API and AI Studio.
Scene extension lets a generated video continue beyond its original length, supporting longer narrative sequences instead of isolated short clips.
Specifying first and last frames gives developers direct control over how a generated sequence begins and ends, rather than leaving the full arc to the model's own interpretation.
Drafting in 360p first, then upscaling the chosen result to 4K, separates the cheap iteration step from the expensive final render.

Drafting at 360p before committing to a 4K render is the useful part here, since it lets a developer test ten variations of a scene for what one full-resolution attempt would normally cost. Specifying first and last frames separately also means the model is filling in a defined gap instead of guessing at an entire sequence from a single prompt.
AI TOOLS
Higgsfield launched FLUX.3 Video Upscale, native upscaling built for natural motion and realism at high resolutions, supporting 2K and 4K output with both Precise and Creative modes for cleaner textures and finer detail.
Precise and Creative modes give two distinct upscaling behaviors, one prioritizing fidelity to the source footage, the other allowing more generative reinterpretation of fine detail.
Native support for motion specifically targets the artifacts and warping that typically show up when a general image upscaler is applied frame by frame to video instead.
Output resolution scales up to full 4K, positioning this for footage that needs to hold up on a large screen, not just a quick social clip.

Precise mode keeps upscaled footage close to the original, which matters if you're delivering client work where invented detail isn't acceptable. Creative mode does the opposite, adding texture and detail the source never had, which only makes sense for footage you're willing to let the model reinterpret.

HOW TO AI
Claude Design generates websites, slide decks, and short videos from a single prompt, included free on any paid Claude plan. The real gap between "AI-generated" and "actually looks designed" is feeding it a documented brand system first, instead of letting it guess your visual identity from a text description.

Step 1: Get access and test it
Claude Design unlocks automatically on Pro or Max at claude.ai/design. Run a quick test: a landing page, then try converting it into a slide deck or short video to see the range.
Step 2: Turn a video into better slides
Asking directly for slides tends to produce something serviceable. Asking for a video first, then converting that into slides, forces a clearer visual narrative before anything locks into a static deck:
Turn this video into a slide deck, one slide per scene, keeping the same visual style and narrative orderStep 3: Feed it your brand first
This is the actual difference between generated and designed output. DESIGN.md covers visual theme, a color palette mapped to roles, typography, spacing, and an explicit "avoid" list, that last part alone produces a bigger accuracy jump than any amount of detail on what to do.
Look through my past marketing materials, my website, and my logo files. Build me a DESIGN.md file documenting my actual color palette with hex codes mapped to roles, typography, spacing, component rules, tone of voice, and a clear list of things to avoid
Step 4: Copy any brand's design system
UIDrop, a free Chrome extension, reads a page's live rendered DOM, colors, fonts, spacing, component styles, from literally any website. Snap a page, send it to Claude, and build with that visual system using entirely different content.
Step 5: Check it, then design without leaving Claude Code
Before shipping, run a quick check:
Check this page for accessibility issues, generate a mobile and tablet responsive version, and give me two alternate headline and CTA options for A/B testingIf already in Claude Code, /design opens a native artboard directly in the terminal, no browser tab needed.

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

ChatGPT added the ability to personalize a Temporary chat with existing memories, custom instructions, and plugins, and save it to history afterward if it turns out to be worth keeping.
Nous Research launched real-profile browsing for Hermes Agent, letting it act from a managed copy of a user's existing Chrome profile, complete with real logins already in place.
Google introduced Expert Intelligence in Gemini Notebook, a cross-Google initiative starting with eligible Google Play ebooks, letting users combine expertise from favorite authors with other sources.

π¬ Krea: H3 Max now live, 50x faster than any other high-quality video model, generating 15 seconds of video in just 5.
Β π Cohere: Parse 5, its newest document parsing model with the strongest price-performance on the market, built for high-volume enterprise work.

THATβS 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
β Full archive of premium guides with ready-to-use prompts
β Structured AI courses (step-by-step, start-to-finish)
β Every upcoming premium tutorial




