OpenAI’s most recent Dev Day took things to a whole new level.
Sam Altman walked on stage with a lineup so dense that even seasoned developers were struggling to keep up.
But don’t worry — let’s unpack it together, one exciting update at a time.
Table of Contents
ChatGPT Now Talks to Apps
Let’s start with what most users will notice first.
ChatGPT can now interact directly with apps — right inside the chat window.
Imagine typing something like:
“Coursera, teach me machine learning.”
And boom — ChatGPT connects with Coursera, pulling up personalized learning content without you ever leaving the chat.
It doesn’t stop there.
ChatGPT can now suggest apps automatically during your conversation.
See also: Mastering the Linux Command Line — Your Complete Free Training Guide
If you mention you’re planning a trip, it might recommend Expedia or Booking.com — and help you search for flights or hotels right away.
Currently, integrated apps include Canva, Coursera, Figma, Spotify, Zillow, Expedia, and Booking.com.
More partners are expected to join by the end of the year.
This marks a big shift. ChatGPT isn’t just a chatbot anymore — it’s becoming an AI-powered workspace that connects your favorite tools in one place.
A New Playground for Developers: Apps SDK
To support this new world of in-chat apps, OpenAI also launched the Apps SDK.
It’s built on the Model Context Protocol (MCP) — an open standard that lets ChatGPT connect securely to external tools and data.
Developers can now build, test, and publish their own ChatGPT apps, complete with custom UIs and interaction logic.
Even better, the SDK is open source.
That means apps built with it can run in any environment that supports MCP — not just ChatGPT.
Later this year, OpenAI plans to open submissions for new apps and introduce monetization details, plus a public app directory so users can explore and install what they need.
Sounds a lot like the App Store moment for ChatGPT, right?
Build Your Own AI Agents with AgentKit
Now, here’s the part developers are buzzing about: AgentKit.
Think of it as a full toolkit for building, deploying, and managing AI agents — without reinventing the wheel.
AgentKit includes three main modules:
- Agent Builder — a visual canvas where you can design your agent’s logic by connecting nodes, tools, and workflows.
- Connector Registry — a centralized dashboard to manage data connections and permissions across platforms like Dropbox, Google Drive, SharePoint, and Teams.
- ChatKit — an easy way to embed chat-based agents into your own app or website, complete with a ready-made interface.
During Dev Day, an OpenAI researcher even built a working agent for the event’s official website — in under eight minutes.
Pretty wild, right?
Smarter Evaluation Tools
OpenAI also revamped its Evals system — the platform developers use to test and measure AI agent performance.
Now it supports:
- Custom datasets for evaluation
- End-to-end “trajectory scoring” to track how an agent performs across an entire workflow
- Automatic prompt optimization based on test results
- Third-party model evaluation right inside the OpenAI ecosystem
All of these tools (including ChatKit and Evals) are now available to all developers at no extra cost.
Codex Is Back — and It’s More Capable Than Ever
Remember Codex, OpenAI’s code-writing model? It’s back — and now officially available.
The new Codex comes with three major upgrades.
1. Slack Integration
You can now @Codex directly inside Slack to ask coding questions or assign tasks.
It automatically grabs the context from your conversation, spins up a temporary environment, and replies with a link to the results.
From there, you can merge, iterate, or continue the work locally.
2. Codex SDK
This SDK lets you embed Codex into your own tools or CI/CD pipelines — using just a few lines of TypeScript.
It supports structured output, context management, and even comes with a GitHub Action for automation.
3. New Admin Controls
Workspace admins can now manage Codex environments, monitor usage and code quality, and set secure local defaults — all through a central dashboard.
Currently, the Slack integration and SDK are available to Plus, Pro, Business, Edu, and Enterprise users.
The new admin tools are rolling out to Business, Edu, and Enterprise plans.
GPT-5 Pro and Beyond
Of course, OpenAI couldn’t leave without unveiling a new model.
Meet GPT-5 Pro — now available via API.
It’s designed for enterprise-level workloads, priced at $15 per million input tokens and $120 per million output tokens.
Yes, it’s pricey — but it’s also the most powerful GPT model yet, outperforming o3-pro in reasoning and context handling.
For developers looking for something lighter, there’s also GPT-Realtime-Mini, a smaller, cheaper version of OpenAI’s real-time voice model.
It offers the same expressive quality but costs about 70% less — perfect for interactive assistants and real-time apps.
And finally, OpenAI teased Sora 2, the next-generation video model that generates synchronized video and audio.
Developers can now control duration, resolution, and aspect ratio — a big step toward fully controllable video generation.
Wrapping Up
This year’s Dev Day wasn’t just about new models.
It was about turning ChatGPT into a platform — a place where apps, agents, and users meet in one conversational space.
From Apps SDK to AgentKit to Codex and GPT-5 Pro, OpenAI is laying the groundwork for a new AI ecosystem — one that feels less like “a chatbot,” and more like a connected, intelligent operating system for work and creativity.
Ready to build your own?
Don’t worry — it’s all easier than it sounds.
👉 Official link: openai.com/devday
