OpenAI just made a simple move with huge consequences: it put an app directory inside ChatGPT and told developers to start shipping chat native experiences. You can browse apps from the tools menu, connect them, and trigger them by name with the @ mention flow. 

This is the real point: ChatGPT is trying to become the place where intent starts and finishes. Not “help me write a list.” More like “help me decide, then actually do it.” That is why OpenAI is pushing the Apps SDK in beta and opening submissions, because a platform without developers is just a feature.

The apps that matter are the ones that collapse a full workflow into a chat. Grocery is the easiest example. With the DoorDash app inside ChatGPT, the pitch is turning recipes and ideas into an order you can actually place. Instacart is going even harder with Instant Checkout inside its ChatGPT app, which is basically a test of whether chat can become the store. We also can’t forget about the Apple Music integration with ChatGPT.

Shopping is next. Target is openly positioning this as holiday muscle, with its own Target app experience in ChatGPT built around curated suggestions and quick purchase paths. Travel is also getting pulled into the same gravity. Tripadvisor now treats this as a front door, telling users to find its app in ChatGPT so reviews and planning live inside the conversation.

Apps only work if they can see enough context to act, and that means privacy and permissions are the whole game. OpenAI says app submissions must include privacy policies, and users can disconnect apps at any time. That sounds clean on paper. The messy part will be how comfortable people feel connecting services that hold their money, travel, inbox, and purchases to a single assistant UI.

Also, availability is not universal. OpenAI notes that access and features vary by plan and region, and some regions have limits. So if you do not see the directory yet, it might not be you.

Text to video is fun, but videos you can actually direct is money. Luma’s new Ray3 Modify release is built around a simple promise: keep the original performance, then change the world around it. That means the timing, eyeline, and emotion from real footage can stay intact while you swap wardrobe, lighting, environment, or even character identity.

The control knobs are what make it different. Luma is leaning on keyframes and character references, with Start and End Frame controls that guide transitions, continuity, and blocking without forcing you to brute force prompt your way through a scene. If you have ever watched an AI video drift off model between frames, you already understand why this matters.

This is not just about generating a movie. It is more like, shoot once, then iterate fast. That is a production workflow studios, agencies, and brand teams can actually use, because it keeps humans in the loop and turns AI into post production leverage.

Coverage from around the industry has framed it as a way to modify footage using reference images while preserving performance, plus generating transitions between two frames when you need the connective tissue.

The tradeoff is that control still has limits. The model can only respect what you give it. Bad source footage, weird motion blur, or complex interactions can still break realism. But this direction is clearly the next phase: fewer magic prompts, more editable levers.

Luma is also scaling like it expects to be a major frontier player. The company says it raised $900M in a Series C led by Humain and tied that to a plan for massive compute via a 2 gigawatt AI supercluster partnership.

Luma AI may not be as well-known as RunwayML, Google or OpenAI, but they are making a lot of noise while pushing the AI video generation industry forward.

🧬 AI tools are helping those with long COVID

💸 ByteDance is spending big to secure AI chips 

⛓️‍💥 Cybercriminals leveraging AI to trick people into clicking unsafe links

👶 A Florida bill targets AI chatbots for minors

🔎 Google blocks AI training on its search results

⌚ AI wearable improves preventative care for seniors

🎁 Step 1: Pick three constraints

Give ChatGPT your rules in one message: budget, time limit, and who is in the room. Add one “no phones” window to make it real.

🍪 Step 2: Turn one idea into action

Have ChatGPT generate two menu options, then use your grocery app choice to turn the winning option into a cart. If you want zero friction, choose delivery over a store run.

🎶 Step 3: Lock the vibe

Ask for a playlist plan based on the group’s ages and energy, then push it into your music app so it is not just text.

🧩 Step 4: Add one interactive moment

Have ChatGPT create a 12 question family quiz that includes two questions per person, based on memories everyone can verify. Keep it fast, keep it funny, and end with one “future plan” question.

📸 Step 5: Make a keepsake

Ask for a one page recap you can paste into a group chat tomorrow morning: what you did, what surprised you, and one photo prompt for next year’s card.

Will ChatGPT’s app store ever rival Google Play or Apple? And, how much more realistic will AI videos get? Happy holidays from Jumble! 🎁

Stay informed, stay curious, and stay ahead with Jumble!

Zoe from Jumble