Open AI had their dev day today where they launched a series of initiatives. If you watched the dev day, you may have taken notice to
Apps on ChatGPT
AgentKit, ChatKit and a Visual Agent builder canvas
GPT 5 Pro & Sora 2 via API
Key Implications
OpenAI has restarted its attempts to create an app ecosystem around chatGPT another attempt at kickstarting the GPT economy
Launched an AgentKit & ChatKit to potentially compete with n8n and zapier with a chatbot like twist
The Apps Gambit on ChatGPT - Round 2
The plugin store did not quite work out for OpenAI facing challenges with adoption, brittleness of capabilities. This seems to be a new attempt to build on top of an MCP like ecosystem to connect apps with native experiences within chatGPT. The use cases shown were figma, canva and zillow. It does make sense that developers building MCPs for their apps be allowed to have a bit more control over the front end via chatGPT. I think this will be more useful for B2C like applications because they tend to have simpler flows. I can’t imagine someone trying to chat their way through designing a poster on Canva so it does seem like the use cases that they picked were off. But having distribution on chatGPT could be a major unlock for b2c companies with more transactional flows like Uber, Doordash or Instacart (hello Fidji, new apps CEO, old instacart CEO).
The Token Economy
OpenAI also shared a list of their developers along with the tiers of Tokens processed. This is really interesting because it helps us understand the $3Bn in spend that is expected to happen in 2025 which is about 20-30% of openAI’s revenue.
They revealed three categories -
Tier 1 - 1 Trillion Tokens
Tier 2 - 100B Tokens
Tier 3 - 10B Tokens
This also gives an interesting insight into what kind of applications are consuming more tokens on OpenAI. After some AI prompting and stringing together multiple tools, I was able to procure the list of companies and here are the key insights.
OpenAI dev day tokens processed by developers
Distribution & Concentration
The allocation to different categories is pretty balanced with public companies or unicorns at 34.8% and AI startups at 32.3%. Major tech companies are well-represented: NVIDIA, Adobe, Oracle, Shopify, Netflix, Target. Financial services companies include Citi, UBS, Stripe, and fintech players like Upstart. Duolingo and Notion are one of the biggest consumers in the new age internet companies category.
AgentKit & the Automation Stack Invasion
Along with the new Apps, OpenAI also announced:
Agent Builder - A visual canvas for multi-agent workflows
ChatKit - An embedded chat interface with customisation of branding, etc.
Connectory Registry - A place for admins to manage tools and connections across OpenAI products.
Evals - way to upload datasets for testing and automated prompt optimisation.
While there are a lot of people that feel that OpenAI is taking aim at Lindy, n8n and other node based workflow builders, there are some important distinctions between them.
OpenAI’s focus seems to be in the arena that they play an excellent game on - chat fist experiences with slick interfaces and interactions. In case you have ever built one of these apps you’ll know the challenges with streaming tokens and tool calls. ChatGPT solves this out of the box for you.
You can build your own chatbot, customize the front end and have it hooked up to a workflow planner where you can configure the flow of actions and tool calls based on different scenarios.
It is missing a comprehensive trigger system that most automation platforms have which can be run manually or via webhooks.
The only way you can trigger a workflow is via a chatbot.
So this feels like a way to create an advances ChatGPT like chatbot that can trigger tools and external services with customized logical flows.
References
OpenAI DevDay 2025 Official Page
https://openai.com/devday/Announcing DevDay 2025
https://openai.com/index/announcing-devday-2025/Introducing AgentKit
https://openai.com/index/introducing-agentkit/Apps in ChatGPT SDK Announcement
https://openai.com/devday/ (Apps SDK section)OpenAI launches AgentKit to help developers build and ship AI agents
https://techcrunch.com/2025/10/06/openai-launches-agentkit-to-help-developers-build-and-ship-ai-agents/OpenAI launches apps inside of ChatGPT
https://techcrunch.com/2025/10/06/openai-launches-apps-inside-of-chatgpt/Sam Altman says ChatGPT has hit 800M weekly active users
https://techcrunch.com/2025/10/06/sam-altman-says-chatgpt-has-hit-800m-weekly-active-users/OpenAI DevDay 2025: Live updates and analysis
https://www.cnbc.com/2025/10/06/open-ai-devday-live-updates-altman-jony-ive.htmlOpenAI is gearing up to release Agent Builder during DevDay
https://www.testingcatalog.com/openai-prepares-to-release-agent-builder-during-devday-on-october-6/
FAQs: OpenAI Dev Day 2025 - Apps, AgentKit & the Token Economy
General Questions
Q: What did OpenAI announce at Dev Day?
A: Three main things:
Apps on ChatGPT - native app integrations within ChatGPT
AgentKit & ChatKit - developer tools for building agents and chat interfaces
GPT-5 Pro & Sora 2 via API - their latest models available through the API
Q: When was Dev Day 2025?
A: October 6, 2025 in San Francisco.
Apps on ChatGPT
Q: Didn't OpenAI already try this with the plugin store?
A: Yep. The plugin store didn't quite work out—had issues with adoption and the capabilities were pretty brittle. This is round 2, building on top of an MCP-like ecosystem to give developers more control over the frontend experience.
Q: Which apps are launching on ChatGPT?
A: So far we've seen Figma, Canva, and Zillow. The real opportunity is for transactional B2C companies like Uber, DoorDash, and Instacart (hello Fidji, new Apps CEO, old Instacart CEO).
Q: What types of apps actually make sense on ChatGPT?
A: B2C apps with simple, transactional flows. I can't imagine someone trying to chat their way through designing a poster on Canva, so the use cases they picked were a bit off. But ordering food, booking a ride, making a reservation? Those could work well with ChatGPT's distribution.
The Token Economy
Q: What's this token economy data OpenAI shared?
A: OpenAI revealed how much their developers are using, broken into tiers:
Tier 1: 1 Trillion+ tokens
Tier 2: 100B tokens
Tier 3: 10B tokens
This is really interesting because it helps us understand the $3Bn in spend expected in 2025, which is about 20-30% of OpenAI's revenue.
Q: Who's consuming the most tokens?
A: Pretty balanced distribution across categories:
34.8%: Public companies and unicorns (NVIDIA, Adobe, Oracle, Shopify, Netflix, Target)
32.3%: AI startups
Financial services (Citi, UBS, Stripe, Upstart)
New age internet companies (Duolingo and Notion are some of the biggest consumers)
AgentKit & ChatKit
Q: What exactly is AgentKit?
A: It's a toolkit with four components:
Agent Builder - visual canvas for multi-agent workflows
ChatKit - embeddable chat interface you can customize
Connector Registry - place for admins to manage tools and connections
Evals - upload datasets for testing and automated prompt optimization
Q: Is this OpenAI taking aim at Zapier and n8n?
A: A lot of people feel that way, but there are important distinctions. OpenAI's focus seems to be in the arena where they play an excellent game—chat-first experiences with slick interfaces and interactions. If you've ever built one of these apps, you'll know the challenges with streaming tokens and tool calls. ChatKit solves this out of the box for you.
Q: What's the big limitation vs traditional automation platforms?
A: It's missing a comprehensive trigger system that most automation platforms have, which can be run manually or via webhooks. The only way you can trigger a workflow is via a chatbot.
Technical Stuff
Q: What's MCP?
A: Model Context Protocol—an open standard that lets apps connect with AI systems. It's what OpenAI is building on top of for this new apps ecosystem, giving developers more control over the frontend via ChatGPT.
Q: Can I use Agent Builder like Zapier?
A: Not really. You can build your own chatbot, customize the front end, and have it hooked up to a workflow planner where you can configure the flow of actions and tool calls based on different scenarios. But without proper triggers, it's not a replacement for traditional workflow automation.