From the very beginning, when we first integrated artificial intelligence into our online courses and learning tools, we chose to build on the OpenAI/ChatGPT API. Over time, other platforms like Anthropic’s Claude and Google’s Gemini have also become valuable options for educators. We recently explored many of these tools in our blog post introducing our free course, Boosting Productivity with AI, which helps teachers learn how to use Custom GPTs, Gems, and other emerging AI tools effectively. Still, our core infrastructure, especially for Verónica (our AI tutor) and our scenario-based learning modules, continues to evolve on OpenAI’s foundation.

That’s why today’s announcements at OpenAI Dev Day 2025 feel so exciting. They open up new doors for more immersive, interactive, and personalized learning experiences. We’re already mapping out how to bring these tools into our learning ecosystem, strengthening what we’ve built and pushing our scenario-based activities to the next level. Here’s what stood out (from our perspective), and how we’re thinking of putting these into action.

New AgentKit

One of the biggest highlights is AgentKit: a toolkit and framework to make it easier to build, deploy, and manage AI agents. 

As a developer and instructional designer, one of the announcements that caught my attention the most was AgentKit. It represents a major step forward in how we can design and organize intelligent learning experiences. Rather than having to hard-code every step of a learning sequence, AgentKit introduces a visual environment where we can map out how an agent thinks, responds, and connects with other systems. This means that the multi-step interactions we’ve been building (like branching dialogues, decision-making moments, and adaptive feedback loops) can now be modularized and managed in a cleaner, more efficient way.

What’s most exciting is that AgentKit isn’t just about convenience; it’s about possibility. It opens the door for creating agents that truly guide learners through a process (asking questions, adapting to their answers, and providing tailored feedback) while maintaining control over structure and quality. With its integrated tools for connecting to databases and external systems, we can imagine developing specialized versions for universities or districts, each customized to their needs but built on the same robust foundation.

Of course, this is something we’ll need to prototype carefully. We want to make sure that our agents behave responsibly, that the logic flows naturally, and that the overall learner experience remains seamless. But what excites me most is that this technology removes much of the complexity behind the scenes, allowing us to focus more on designing meaningful learning experiences and less on the technical overhead of managing them.

Sora API

Another development that immediately sparked my imagination is Sora, OpenAI’s text-to-video generation system, which is now accessible through the API. For those of us working on scenario-based learning, this feels like the missing piece we’ve been waiting for. Until now, most of our interactive activities have relied on text and audio, or on pre-designed videos created with other AI tools, to simulate real-life situations. With Sora, we can begin to visualize those same scenarios, turning dialogues and prompts into short, dynamic videos that add depth, movement, and emotion.

As an instructional designer, I see enormous potential here. Imagine a language learner not just listening to a conversation, but actually seeing it unfold: the gestures, the facial expressions, the setting, the cultural context, all those subtle cues that make communication authentic. In bilingual and teacher preparation programs, this could be transformative. We can adapt the videos to a learner’s level or teaching focus, and even generate variations on the fly to keep practice fresh and meaningful.

Of course, this is still early technology, and we’ll need to test it carefully. We’ll experiment with quality, character consistency, timing, and the balance between creativity and clarity. Our goal isn’t to replace what already works, but to enrich it, to bring a layer of realism that helps learners connect emotionally and cognitively to what they’re practicing. Once we find the right rhythm and parameters, I believe Sora will allow us to create scenario-based learning experiences that feel truly immersive and alive.

GPT-5 Pro in the API

Another announcement that immediately stood out to me was the release of GPT-5 Pro in the API. We’ve been working with GPT-5 for a while now, and it’s already proven to be a remarkably capable model for building adaptive learning experiences. But this new version brings a level of power and flexibility that can fundamentally change how we design complex educational interactions.

What makes GPT-5 Pro so promising is its expanded reasoning capacity and ability to process more information in parallel. In practice, that means we can create activities that involve multiple steps of thinking (analyzing, interpreting, reflecting) without losing coherence or precision. For example, in a scenario where a learner must analyze a text, defend an opinion, and then critique their own argument, the model can now manage those transitions more smoothly, providing layered and meaningful feedback along the way.

For us as instructional designers, this opens the door to richer, more authentic cognitive tasks. It gives us the ability to embed deeper reasoning directly into the learning flow, rather than simplifying it for technical constraints. Of course, we’ll continue to test how the Pro model performs in different contexts. However, the fact that this level of reasoning is now accessible through the API means we can start experimenting immediately, refining how our AI tutor and scenario-based activities guide learners through increasingly sophisticated challenges.

Other Announcements

From a development perspective, a few other updates from Dev Day will also make a tangible difference in how quickly we can bring new ideas to life. The release of Codex, now generally available, gives us a more stable foundation for coding and automating the technical side of our projects. Whether we’re generating scripts, integrating APIs, or refining user interactions, Codex will help us streamline the process and experiment more freely without starting from scratch each time.

On the interaction side, the new GPT Realtime Mini model opens exciting possibilities for voice-based learning. Its speed and responsiveness make it easier to design activities where learners can speak, listen, and receive feedback in real time, which is something that has always been central to our vision for Verónica and our speaking practice modules. Similarly, the introduction of GPT Image 1 Mini expands what we can do with visuals, allowing us to generate quick, context-specific images to support comprehension or illustrate scenarios within a lesson. We see particular potential for this in our work with the Bilingual Homework Hotline, where pre-service teachers provide homework support to emerging bilingual students via Zoom. In that setting, being able to generate clear and timely images can make explanations more accessible, helping students better grasp complex concepts in real time. Together, these tools bring us closer to creating experiences that feel immediate and dynamic, less like static content and more like living, evolving learning environments.

Stay Tuned

With all the announcements from OpenAI Dev Day 2025, we’re once again excited about the possibilities these new tools open for education. They give us the opportunity to create richer and more immersive activities, ones that are engaging for pre-service and in-service teachers, and more responsive and personalized for every learner. By blending video, voice, and real-time interaction, we’ll be able to design experiences that feel more alive, meaningful, and connected to authentic classroom practice.

For those who haven’t read our earlier post on teacher preparation using AI, these advancements also mean that we can now design entire learning modules or online courses tailored to institutions. We can create safe spaces where pre-service and in-service teachers can explore, experiment, and interact with realistic classroom scenarios, remembering that these systems are not meant to replace human guidance but to amplify it. Each environment we create is built to keep the human in the loop: a mentor, professor, or instructional coach who can support teachers as they reflect, practice, and grow. Whether it’s a student teacher preparing for certification, or a new educator navigating the challenges of a bilingual or ESL classroom, our goal is to use AI to strengthen (not replace) the human connection in learning.

We truly believe that this technology has the potential to transform education, but its success depends on us: the teachers, instructional designers, and education professionals who understand how learning occurs. Our role is to shape these innovations with purpose, to use them thoughtfully, and to build experiences that enhance understanding rather than simply automate it. The next generation of learning will not be built by technology alone; it will be built by educators who know how to use it well. If you’re interested in learning more about how we can bring AI-powered scenario-based learning and practice activities to your pre-service or in-service teachers, please feel free to contact us. We would love to collaborate.

If you’d like to watch the full presentation from OpenAI Dev Day 2025, you can find the video below.