Netflix built this insane AI

Welcome back to The Shift. Let’s get straight to what matters in AI today…

Today we have:

🎥 Netflix Built a Tool That Erases Video Objects and Rewrites the Physics Around Them

 🏃How to Run Gemma 4 Locally on Your Phone With Google AI Gallery

🧠 Anthropic limits Claude's access for agent tools

🔨Tools and Shifts you cannot miss 

🎥 Netflix Built a Tool That Erases Video Objects and Rewrites the Physics Around Them

Netflix released VOID, an open-source framework that doesn't just remove objects from video; it figures out what should physically change in the rest of the scene once that object is gone.

The Shift:

Current Tools Just Paint Over Gaps - Every existing removal tool inpaints the background behind a deleted object, but none of them reason about what happens next, so if you remove something that was holding or hitting another object, the result looks wrong.

VOID Thinks About Cause and Effect - A vision-language model identifies which parts of the scene are causally affected by the removed object, then a video diffusion model generates physically plausible outcomes like a balloon floating away when you erase the hand holding it.

It Handles Physics It Was Never Trained On - Demos show blocks not falling when a chain link is erased and objects changing trajectory after a collision partner disappears, with a two-pass system that fixes shape morphing artifacts when they appear in the first output.

Evaluators Preferred It Over Runway - 25 evaluators compared VOID against six baseline models, including Runway, and preferred Netflix's results nearly two-thirds of the time, and the entire framework is open source for anyone to use and build on.

Video editing has always been about removing things. VOID is the first tool that actually thinks about what should happen after you remove something. For creators, editors, and production teams, this is a fundamental shift from cosmetic fixes to physics-aware editing.

Together with Mintlify

Your Docs Deserve Better Than You

Hate writing docs? Same.

Mintlify built something clever: swap "github.com" with "mintlify.com" in any public repo URL and get a fully structured, branded documentation site.

Under the hood, AI agents study your codebase before writing a single word. They scrape your README, pull brand colors, analyze your API surface, and build a structural plan first. The result? Docs that actually make sense, not the rambling, contradictory mess most AI generators spit out.

Parallel subagents then write each section simultaneously, slashing generation time nearly in half. A final validation sweep catches broken links and loose ends before you ever see it.

What used to take weeks of painful blank-page staring is now a few minutes of editing something that already exists.

Try it on any open-source project you love. You might be surprised how close to ready it already is.

Try Atlas today!

 🏃How to Run Gemma 4 Locally on Your Phone With Google AI Gallery

Fully offline. Completely private. No API. No subscription. Here's how to get it running.

Step 1: Download Google AI Gallery

• iPhone: App Store

• Android: Google Play

Step 2: Choose your mode. Pick from AI Chat, Ask Image, Audio Scribe, Agent Skills, and more. For a straight conversation with Gemma 4, select AI Chat.

Step 3: Pick your model

• Gemma E4B, more capable, needs 8GB RAM minimum

• Gemma E2B, lighter, still highly capable on most phones

Step 4: Start chatting. That's it. The model runs entirely on your device. Turn on airplane mode, and it still works, no internet, no data, leaving your phone, no limitations.

100% open source. Multimodal across text, audio, and image. Works on both iOS and Android. You can download Google AI Gallery here.

🧠 Anthropic limits Claude access for agent tools

Anthropic announced that Claude subscriptions will no longer support third-party agent platforms like OpenClaw starting tomorrow. The company says rising demand for these tools broke its flat pricing model. Users will now need to pay separately for usage. 

The Shift:

1. Agent tools drove an unexpected usage surge - Agent platforms like OpenClaw were sending continuous high-volume requests to Claude, far beyond typical usage patterns, putting pressure on infrastructure and exposing limits of subscription-based pricing models.

2. New pricing model replaces unlimited access - Users can still access Claude through third-party tools, but only via API keys or paid usage bundles, marking a shift from flat subscriptions to usage-based pricing for agent-driven workflows.

3. Credits, discounts, and refunds introduced - Anthropic is offering one-time credits equal to a monthly subscription, discounted usage bundles up to 30%, and refund options, as it tries to manage backlash and smooth the transition. 

4. Pushback from open-source community - OpenClaw creator Peter Steinberger criticized the move, saying Anthropic is copying open-source innovation while restricting access, raising concerns about control, ecosystem openness, and future developer trust.

This is where AI pricing breaks. Flat subscriptions worked for chat, but agents changed the game. Now the real battle is not just models, it is who controls access when AI starts working nonstop. 

Together with Gladly.ai

88% resolved. 22% loyal. Your stack has a problem.

Those numbers aren't a CX issue — they're a design issue. Gladly's 2026 Customer Expectations Report breaks down exactly where AI-powered service loses customers, and what the architecture of loyalty-driven CX actually looks like.

Get the report

🔨AI Tools for the Shift

🏠 Izbaviz - Instantly visualize and design interiors with AI across multiple room types and styles without needing a designer.

🎮 LHM.gg - Run esports tournaments and broadcasts seamlessly with an all-in-one competitive gaming platform.

📄 WhatsTheCatch - Review and analyze documents using Claude-powered AI for deeper insights and clarity.

✍️ Infinity Copy - Analyze your market, define positioning, and generate high-converting copy backed by proven frameworks in minutes.

🕹️ Chatforce - Create and share viral interactive games powered by AI agents for engaging user experiences.

🚀Quick Shifts

🔐 Mercor, a company that supplies AI training data, suffered a security breach, prompting Meta to pause collaboration and OpenAI to launch an investigation.

⚠️ Microsoft’s Copilot terms state the AI is “for entertainment purposes only,” warning users not to rely on its outputs for important decisions as the company works to update outdated language.

🔄 OpenAI reshuffled its leadership, with COO Brad Lightcap moving to lead “special projects,” while other executives take on expanded roles amid ongoing product and enterprise growth.

🚗 OpenAI rolled out ChatGPT support for Apple CarPlay, enabling hands-free access to Voice Mode for in-car conversations and assistance.

🎥 Pika Labs launched PikaStream 1.0 in beta, enabling AI agents to join Google Meet calls as real-time video avatars with voice cloning and live conversational capabilities.

🧩 Prompt of the Day

Motivation and Barrier Analysis

Understand what drives customers to buy and what stops them, so you can refine messaging and remove friction.

Paste the prompt: Drop this into ChatGPT and fill in your product.

Prompt to paste

Create a motivation and barrier analysis for [Insert product or category]. Include:
Emotional Factors: [List motivations and barriers driven by feelings, desires, or fears]
Functional Factors: [List practical benefits and objections related to price, quality, usability]
Social Factors: [List influences like status, trends, peer validation, or reviews]
Use Case: Improve messaging, address objections, and highlight key purchase drivers to increase conversions.

🤳AI Nugget of the Day

That’s all for today’s edition see you tomorrow as we track down and get you all that matters in the daily AI Shift!

If you loved this edition let us know how much:

Forward it to your pal to give them a daily dose of the shift so they can 👇