OpenAI’s Dev Day marked a significant turning point in AI, introducing powerful new models and developer tools. Key announcements include GPT-4 Turbo for advanced processing, custom GPTs for personalized AI assistants, and AgentKit to build sophisticated AI agents. These updates empower business owners and professionals to leverage AI more deeply for efficiency, content creation, and innovative customer interactions.
Sam Altman kicked off the keynote by putting the platform’s explosive growth into perspective. Since their first DevDay in 2023, the number of developers building on OpenAI has doubled from 2 million to 4 million. More dramatically, ChatGPT’s weekly user base has skyrocketed from 100 million to over 800 million people. But the scale isn’t just about users; it’s about usage. The platform now processes over 6 billion tokens per minute, up from 300 million in 2023—a 20x explosion in computational throughput that underscores the intensity of this new AI-powered economy.
The rest of the event was a cascade of announcements that redefine what’s possible for builders. Here are the four most impactful takeaways that show how OpenAI is collapsing the time it takes to go from a simple idea to a fully functional product.
ChatGPT Isn’t Just a Chatbot Anymore—It’s Becoming an App Platform.
You can now chat with apps in ChatGPT. pic.twitter.com/T9Owi3POim
— OpenAI (@OpenAI) October 6, 2025
The most significant strategic shift announced was the new “Apps SDK,” which allows developers to build interactive, full-stack applications that run directly inside of ChatGPT. This transforms the conversational tool into a full-fledged operating system, giving developers a direct distribution channel to its 800 million weekly users. Critically, the Apps SDK is built on MCP, an open standard, signaling OpenAI’s intention to create a genuine ecosystem rather than a purely proprietary walled garden.
The live demos illustrated just how powerful this new paradigm will be. An integration with Coursera allowed a user to watch a course video, pin it to the screen, and ask ChatGPT questions about the content in real-time. Another demo showed how a user could brainstorm ideas for a dog-walking business and then simply ask Canva to generate posters and a full pitch deck without ever leaving the conversation. A Zillow app allowed a user to search for homes on an interactive map, filter the results, and ask follow-up questions about proximity to amenities like dog parks.
This isn’t just about embedding web pages in a chat window. It’s a new model for app discovery and user interaction where relevant, powerful tools can be suggested and invoked contextually, right when a user needs them most.
You Can Now Build and Deploy a Custom AI Agent in Under 8 Minutes.
🚨Big news: @OpenAI just launched Agent Kit, and it’s not just another tool.
— Rakesh Gohel 🇨🇦 (@rakeshgohel01) October 6, 2025
It could redefine how businesses design, deploy, and scale AI workflows.
This might be the moment AI moves from tool to infrastructure. 👇 pic.twitter.com/QcfJ0BoVQb
While the excitement around AI agents is high, Sam Altman noted that very few make it out of the prototype phase and into production. The complexity of orchestration, evaluation, and UI development is a major barrier.
BUT FOR ALL THE EXCITEMENT AROUND AGENTS, VERY FEW ACTUALLY MAKE IT INTO PRODUCTION. AND MAJOR USE. IT’S HARD TO KNOW WHERE TO START. WHAT FRAMEWORKS TO TRUST. AND IT’S A LOT OF WORK.
To solve this, OpenAI launched AgentKit, a complete toolkit designed to take agents from idea to production with radically less friction. It includes a visual “Agent Builder,” an embeddable “ChatKit” UI, and a suite of “Evals for Agents.” The true showstopper was watching a production-grade AI agent, complete with custom branding, data sources, and PII guardrails, be visually assembled and deployed to a live website in less time than it takes to make coffee. The agent, humorously named “Ask Froge” and programmed to “speak in the style of Froge” and “ribbit,” went from a blank canvas to a live, interactive tool for thousands of attendees in under eight minutes, a process that would have taken a dedicated team weeks just a year ago.
AI Is Now a Teammate That Can Go from a Sketch to Controlling Hardware.
OpenAI announced that Codex, its software engineering agent, is now generally available and supercharged by a new, purpose-built GPT5-CODEX model. The capabilities demonstrated were nothing short of mind-bending, showing that AI is evolving from a coding assistant into an engineering teammate that can break out of the screen.
This wasn’t just about writing code; it was about AI orchestrating the physical world in real-time. Starting with a mere hand-drawn sketch, Codex was commanded—first with text, then with an Xbox controller, and finally with natural voice—to manipulate a professional camera and the entire auditorium’s lighting rig. He orchestrated this entire sequence without writing a single line of traditional software code himself. The most stunning display of agentic behavior came when Codex, tasked with integrating the room’s lighting system, realized it was missing information. It autonomously went to GitHub, found the necessary documentation for the lighting protocol, and incorporated that new knowledge to complete the task—a clear demonstration of an agent that doesn’t just execute, but actively problem-solves.
This isn’t just a flashy demo; it reflects real-world productivity gains. OpenAI revealed that its own engineers who use Codex complete 70% more pull requests per week, and internal usage of the tool has increased tenfold since August. This adoption has made GPT5-CODEX one of the fastest-growing models ever, having already served over 40 trillion tokens since its release. The productivity gains are staggering, but the qualitative shift is even more profound. As Romain noted, Codex is becoming a tireless teammate, capable of working “for more than 7 hours straight on big refactorings”—a task that would exhaust a human engineer. This isn’t just assistance; it’s superhuman persistence.
Sora 2 Can Now Generate Realistic Video and Synchronized Sound.
Sora 2 Remix → Mario's Escape.
— Rory Flynn (@Ror_Fly) October 3, 2025
The remix feature is underrated.
PROCESS:
1. Generate an intial video
2. Post it.
3. Select Remix.
4. Describe next scene / repeat.
Initial Prompt:
Realistic body cam footage of a police officer pulling over Super Mario in his mario cart. It was… pic.twitter.com/Sn3VwuiGSM
The preview of the Sora 2 API is now available to developers, and it comes with a game-changing new capability. The model can now generate not just stunningly realistic video, but also pair it with “rich soundscapes, ambient audio, and synchronized effects that feel grounded in what you’re seeing.”
This integrated audio-visual generation represents a massive leap forward. The new model’s controllability is a leap forward. For instance, a developer can now feed it a tight, iPhone-style vertical video and prompt Sora to seamlessly expand the scene into a “sweeping, cinematic wide shot,” fundamentally changing the creative possibilities for filmmakers. Another powerful feature is the ability to start with a single still image—like a photo of a pet dog—and have Sora 2 generate a complete, dynamic video scene around it.
By bundling realistic video with a fully integrated soundtrack, OpenAI is providing a tool that moves far beyond simple video clips and toward the creation of fully immersive and believable AI-generated media.
Conclusion: From Idea to Reality in Minutes
The unifying theme of DevDay 2025 was the dramatic compression of the creation lifecycle. Tools that were once complex and accessible only to specialized teams are now becoming simple, powerful, and available to anyone with an idea. As Sam Altman stated in his closing remarks, the barriers to building are collapsing at an unprecedented rate.
SOFTWARE USED TO TAKE MONTHS OR YEARS TO BUILD. YOU SAW TODAY. IT TAKES MINUTES NOW. AND TO BUILD WITH AI YOU DON’T NEED A HUGE TEAM. YOU DON’T NEED A BUNCH OF INFRASTRUCTURE. YOU JUST NEED A GOOD IDEA.
With the time from concept to product shrinking from years to minutes, the landscape for developers, creators, and entrepreneurs has been fundamentally altered. The only question left is: with the barriers to creation falling this fast, what idea will you bring to life?
