The air at Dev Day 2025 was electric. Everyone was waiting for the "iPhone moment." The rumors were a tech enthusiast's fever dream: GPT-6 with persistent memory and personality. A voice assistant that could call the pizzeria for you. Sora-2, unleashed for public use. Jony Ive's sleek AI hardware. A new browser, a new search engine, a built-in store for AI apps. We were braced for ChatGPT to be declared the new operating system.

When Sam Altman took the stage, the tension was palpable, reminiscent of a Steve Jobs keynote in 2010. But as the presentation unfolded, half of our expectations dissolved into fantasy. GPT-5 arrived, but without memory. The advanced voice agent was still in beta. No hardware was revealed.

Instead, OpenAI delivered something that, at first, nobody fully understood. They gave us the thing we weren't watching for. ChatGPT was no longer a product. It was now a platform.

The Platform Is the Message

This wasn't an upgrade; it was an annexation. OpenAI unveiled a series of tools that transform its chat window into a central hub for digital life, moving from a destination to a true operating environment.

1. The Apps SDK: An App Store in a Chat Window

OpenAI rolled out its Apps SDK, a toolkit for developers to build applications that run natively inside ChatGPT. The first wave of partners is already here: Canva, Figma, Spotify, and Zillow. You don't get sent to another tab; the app's functionality opens right in your conversation. Later, a full App Store will follow, allowing developers to monetize their creations. This isn't just an integration; it's the beginning of a new application economy.

2. AgentKit: The Visual AI Assembly Line

Alongside the SDK, OpenAI launched AgentKit—a visual, no-code editor for building your own AI agents. In your browser, you can drag and drop logic, connect to APIs, define actions, and run automated tests. What used to be a two-month development project for a small team can now be assembled in two evenings by a single person. These agents can be embedded in your own products or run directly within the new ChatGPT platform. If you're looking for productivity gains, our guide to top AI productivity agents provides more context on this shift.

3. Sora and GPT-5 Pro: The Engine Upgrade

And what powers this new world? Sora is now fully integrated for Plus subscribers. You simply write a prompt and get a video. And GPT-5 Pro is now open for access. It’s a specialized reasoning model, accompanied by a cheaper mini-version for voice and a GPT-5-Codex with Slack integration. This isn't just a new set of features; it's the powerful, multi-modal engine required to run the new environment.

The Verdict: A Code Red for the Digital Economy

This convergence of a browser, an automation layer, and a generative engine into a single window is an extinction-level event for many established tech categories.

For Businesses & Developers

The barrier to entry for creating powerful AI applications has just been obliterated. You can now build, test, and deploy AI agents and apps with minimal code and no data science team. The project that would have taken two months and a dedicated team can now be accomplished in a weekend. An entire generation of AI-native startups will be born from this.

For Investors

OpenAI has just declared war on multiple trillion-dollar markets simultaneously.

• Automation Services (Zapier, Make): Their market cap ($5.3B for Zapier) seems fragile when ChatGPT can perform the same multi-step automations as a native function.
• App Stores (Apple, Google): With a turnover of $643 billion, the App Store is a titan. OpenAI is building a direct competitor for where users discover, pay for, and use digital services.
• Browsers (Google Chrome): Chrome has 3 billion users because it's the entry point to the internet. If that entry point shifts to a chat interface, the browser becomes a background utility.

For Everyone Else

The way we use the internet is about to fundamentally change. The era of switching between dozens of tabs and apps for different tasks is ending. Your design tool, music player, CRM, and email newsletter will all be accessible from a single window. The world without a browser isn't ten years away. It arrived today.

🚨 The Tsunami Warning

This isn’t a gentle tide. This is a seismic shift. Here's who should be worried:

1. SaaS Companies (Impact Score: 9/10)
If your application cannot be called from within ChatGPT, you are now outside the main current of user flow and money. The user interface is moving.

• What to do: Start building for compatibility now. Get on the SDK. If you’re not in the chat, you risk becoming invisible.

2. Automation-as-a-Service (Impact Score: 8/10)
Zapier, Make, and n8n suddenly look like heavyweight, complex systems from a bygone era.

• What to do: Pivot to hardcore, enterprise-grade B2B integrations and complex scenarios where a generalist model like GPT-5 Pro will fail. The consumer and prosumer market is no longer safe ground.

3. Web Browsers (Impact Score: 7/10)
If you lose the battle for the internet's starting page, you're left with nothing but the rendering engine.

• What to do: Invest everything in deep AI integrations or accept your fate as a background utility. The browser wars are over, and the platform war has begun.

This wasn't the Dev Day we expected, but it was the one that will be remembered. OpenAI didn't just show us a new AI; it showed us a new world for AI to live in.

Let's look at what happened in detail

OpenAI DevDay 3: Major Announcements and New Developer Tools

OpenAI held its third DevDay conference, unveiling significant updates for developers — from powerful new models to revolutionary tools for building agents and applications.

OpenAI's Journey: From 2015 Vision to Reality

Greg Brockman opened the conference by reflecting on OpenAI's origins. In 2015, the founding team created a simple three-step plan:

1. Solve reinforcement learning
2. Solve unsupervised learning
3. Gradually learn really complicated things

This vision materialized through key milestones: the 2017 Dota 2 success demonstrated reinforcement learning at scale, while the "unsupervised sentiment neuron" paper showed that next-token prediction could learn semantics. By 2019, GPT-3 proved useful enough to build products around, leading to the API launch that let developers connect AI to real-world applications.

New Models: Intelligence Across Modalities

GPT-5 Pro

GPT-5 Pro is now available in the API — the same model powering ChatGPT. It's more expensive and thinks longer, making it ideal for use cases where precision is critical.

Key principles for working with GPT-5:

• Keep prompts simple and clear
• Use evals extensively
• Iterate systematically, editing one variable at a time

Sora 2: Video Generation

Two Sora models launched in the API: Sora 2 for fast experimentation, and Sora 2 Pro which pays closer attention to detail for production-ready videos. The technology enables consistent character rendering and complete vibe transformations through re-prompting.

Cost-Efficient Models

• Mini Realtime Model: 70% less expensive than the large model with equivalent voice quality
• New Image Generation Model: 80% price reduction while maintaining quality comparable to the original model

GPT-4o Mini: Open Source Success

The open models achieved 23 million downloads on Hugging Face within months, with hackathon submissions from developers across 147 countries. Use cases range from local security operations centers to emergency mesh networks for natural disasters.

Codex: Transforming Software Development

Multi-Platform Integration

Codex now works across terminal, IDE, GitHub, and the cloud, powered by GPT-5 Codex — a version optimized for agentic coding with adaptive thinking capabilities.

New Enterprise Features

Slack Integration
Developers can assign tasks to Codex directly within team conversations. Team members can mention Codex in threads, and it will create tasks, complete work, and provide screenshots of results.

Multimodal Capabilities
Codex's "best of n" feature uses vision capabilities to generate multiple design variations from low-fidelity sketches, which can then be pulled into VS Code for iteration.

MCP (Model Context Protocol) Support
Codex now supports MCP tools in both CLI and IDE extensions, including integrations with Figma for design-to-code workflows and Chrome DevTools for performance analysis.

GitHub Code Reviews
When activated, Codex automatically reviews every PR, deeply understanding the project and dependencies rather than just performing static analysis. OpenAI uses this internally to catch bugs before production.

Codex SDK

The Codex SDK makes the same agent available in CLI and IDE accessible anywhere developers need a coding agent, enabling integration into custom workflows.

Use cases include:

• Automatic troubleshooting when pagers go off
• Removing tech debt and feature flags post-deployment
• Auto-updating documentation
• Self-evolving applications that can modify themselves

Responses API: The New Standard for Agents

The Responses API is the flagship API for building agents, featuring built-in tools (web search, file search, MCP), support for multiple tool calls in a single request, and preserved reasoning tokens across turns. On average, it's 20% faster than chat completions for multi-turn requests.

Agent Kit: Simplified Agent Development

Agent Kit transforms agent primitives into customizable building blocks, addressing the complexity of orchestrating workflows, finding optimal prompts, deploying UIs, and setting up evals.

Key Components

Agent Builder
Provides out-of-the-box templates with pre-built logic and guardrails, such as customer support agent templates.

Widget Builder
Supports visual widgets to bring conversations to life, including charts, email sending, and purchase capabilities — all matching your brand's look and feel.

Automatic Prompt Improvement
The updated Evals product automatically suggests improved prompts based on completed eval runs. Customers like Carlyle increased accuracy by 30%, while Rippling cut development time by 40%.

Reinforcement Fine-Tuning (RFT) for Agents
RFT now extends to tool calls, enabling models to call the right tools at the right times. Custom graders let teams measure performance on their specific priorities. Cognition found GPT-5 could solve hard coding problems 50% faster with RFT.

Real-World Example: RAMP

RAMP demonstrated a procurement agent built with Agent Kit that reduces software purchasing from weeks to minutes, automating vendor research, policy compliance, and purchase requests.

Apps SDK: Building for ChatGPT

The Apps SDK, available in preview, enables developers to build fully interactive applications inside ChatGPT. Users can chat with apps that respond to natural language, with the SDK built on MCP to sync app servers, models, and UI.

Developer Freedom

• Backend support for Node, Python, or any MCP-supported language
• Frontend flexibility with React or other frameworks
• Full control over the entire chat experience

Key Features

Widget State Management
The Apps SDK introduces widget state, allowing two-way communication between the conversation and UI components. Developers can build visual rich experiences that remain part of the conversation flow.

Personalized Context
ChatGPT's knowledge of user preferences enables apps to provide customized results and recommendations just when users need them.

Visual Learning Experiences
Apps can create immersive, interactive educational experiences directly within ChatGPT.

Launch Timeline

• Now: Developer preview available for building and testing
• Later in 2024: App submissions open for review and publication
• Future: App directory for user discovery and monetization options

DevDay 3 showcased OpenAI's commitment to empowering developers with increasingly sophisticated tools. From multimodal models to autonomous coding agents and integrated ChatGPT applications, the announcements signal a shift toward AI systems that can handle complex, real-world tasks while remaining accessible to developers of all skill levels.

The recurring theme: OpenAI provides the infrastructure and primitives, while developers bring these capabilities to life through practical applications that benefit users worldwide.