OpenAI’s latest DevDay was its biggest yet; not just a demo reel, but a roadmap for how every kind of user will build on ChatGPT. From the debut of Apps in ChatGPT to a new AgentKit, a smaller voice model, and Sora 2 in the API, the platform is officially becoming an ecosystem.

The headline feature is Apps in ChatGPT — small, interactive modules that live directly inside a chat. Users can talk to these apps, trigger them naturally, or browse an “App Store” built around OpenAI’s new Apps SDK (preview).

Partners like Figma, Spotify, Expedia, and Canva already appear as early integrations. For developers, this means interfaces, not just prompts, and for users, it means never leaving ChatGPT to book, build, or browse.

AgentKit provides everything needed to make agents reliable in production — a visual Agent Builder, new evals and trace tools, and a Connector Registry for secure data linking.

It’s meant to replace the patchwork of scripts and API chains many teams rely on today. Enterprise users can now version, audit, and monitor agents like proper software releases.

Sora 2, OpenAI’s most advanced text-to-video model, is now in the API. Developers can generate or remix short, realistic clips, add synchronized sound, and experiment with video-native storytelling. It’s a major leap for creative apps, and early tests show realistic motion, scene continuity, and optional “cameo” support for self-inserted likenesses.

Codex — OpenAI’s code execution model;  is now generally available with Slack integration, a Codex SDK, and new enterprise controls. Teams can build live coding copilots, automate documentation, or query internal tools securely. It’s also deeply linked to the new AgentKit stack for developers who want agents that can reason and run code.

The GPT-5 Pro model is the new flagship for precision-critical work like data cleaning, analytics, or contract review. OpenAI calls it “the smartest model in the API,” emphasizing consistent reasoning rather than creativity. Expect it to anchor enterprise and research use cases over the coming months.

Rounding things out are two new compact models: gpt-realtime-mini, a lightweight voice model that’s 70% cheaper than the full version, and gpt-image-1-mini, a faster, lower-cost image model for instant concept generation and thumbnails. Together, they make real-time and visual AI much more accessible.

Between apps, agents, and new media models, OpenAI is clearly positioning ChatGPT as both a user platform and a developer runtime. What’s next: broader app availability, enterprise-grade connectors, and a wave of startups built entirely inside ChatGPT.

MrBeast said the quiet part out loud. In a new post he asked how creators can compete when AI generated videos become indistinguishable from human work, calling it “scary times.”

His comments arrive as new tools make video generation and remixing vastly easier, including recent launches that added realistic audio and self insertion features. Coverage notes he has tested AI tools himself in the past and faced community blowback when an AI thumbnail generator upset artists, which he later reversed. 

The warning blends practical and ethical risk. If synthetic video can be made at scale and at low cost, ad markets and recommendation systems may favor volume over craft. That pressures original creators on time, budgets, and disclosure.

MrBeast’s track record in operations and optimization makes the warning notable, since he has been quick to adopt tools that increase reach. The core fear is not just copying a style but flooding feeds with convincing clips that are hard to verify in real time. 

Commentary pointed out that large channels like his have experimented with AI workflows and benefitted from automation already, so calls for caution can sound uneven.

But the concern is broadly shared across creator forums as platforms pilot new AI features. The near term fix is clear labelling and consent for training and likeness use. The midterm question is whether platforms will reward provenance, watermarks, and credits in ranking signals so that human craft is discoverable. 

Treat AI video like power tools. Use it for ideation, animatics, or localization, then add human judgement on story, pacing, and ethics. Keep a disclosure template and a credit line for any model or dataset you rely on.

Maintain a human only signature series to protect your brand voice, and monitor policies around synthetic actors and likeness control as these evolve.

🤖 NVIDIA and Fujitsu team up on Japan AI and robotics push

🚗 China’s self-driving firms pivot testing and HQs to Europe

🧱 AMD inks multiyear AI chip pact with OpenAI

🛣️ Tesla starts rolling out FSD v14 with new features

☁️ AWS adds global cross-Region inference for Claude 4.5

🇪🇺 EU drafts Apply AI strategy to cut US China dependence

You don’t need to code — the new OpenAI Agent Builder lets anyone design a working assistant in minutes.

Choose one thing you repeat every week — summarizing articles, organizing notes, finding recipes, checking schedules, or brainstorming posts.

Keep it short and specific: “Summarize meeting notes,” not “Manage my whole job.”

Log in to platform.openai.com/agent-builder.

Click Create Agent, give it a name and description in plain English.
Example:

“TripWhiz helps me find weekend destinations under three hours away.”

Under Knowledge, upload a file or paste key info your agent should reference — a FAQ, a planning doc, a spreadsheet export, or a text list.

You can add multiple small files; the agent automatically indexes them.

Scroll to Behavior → Instructions.

Tell it how to act and what tone to use:
“Be brief, friendly, and always give a next-step suggestion.”
You’ll see the preview update live on the right side.

Toggle the built-in capabilities that fit your task:

Click Test Agent and type a real request it should handle.

Watch how it reasons through the task in the preview.
Tweak its goal or tone until it consistently gives the kind of answer you want.

Click Publish to finalize. You’ll get a link that others can open inside ChatGPT.

Try sending it to a friend — they’ll instantly see your agent running live.

Build one functional, named agent that genuinely saves you time this week — like “MealMate,” 

“Inbox Tamer,” or “TripWhiz.”
Bonus: Post a short caption on social media explaining what it does and tag #AgentBuilder so others can try your creation.

You just shipped your first no-code AI.

Click below ⬇️

Which DevDay features are you the most excited to try first? And, is Mr. Beast right about AI video putting millions of creators’ livelihoods at risk? See you next time! 🚀

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

Zoe from Jumble