OpenAI: The Platform that also wants to be an Aggregator

OpenAI’s recent product launches read like a Platform that also wants to be an Aggregator¹ . Instead of trying to keep everyone inside of ChatGPT, I think they should be more focused on bringing ChatGPT into everywhere else. But I don’t fault them for trying to have it both ways.

The big headline is Monday’s announcement of the ChatGPT Apps SDK, which will allow 3rd party companies to integrate their app into ChatGPT. It’s a pretty naked play at trying to become the mythical Superapp, the device- and operation system-agnostic app from within which all users discover and use other apps² , that so far has only taken off in Asia.

Their core assumption that I disagree with is that people will prefer to use a widget version of Zillow or DoorDash or Expedia embedded within ChatGPT over using those apps directly. Forcing the UX to fit within equivalent of an iframe feels so limiting and greedy, like AOL in the ‘90s. What if, instead, ChatGPT redirected people to the website or app and brought your chat and identity context with it?

The more cynical thought is that this doesn’t actually need to work very well, and that the real goal is to spur tech companies to add “build an agent for ChatGPT” to their roadmaps, and once those companies do that, they’ll naturally move on to embedding ChatGPT into their own app experiences. A win-win for OpenAI.

The Workshop

This is a newsletter-only section where I share a half-baked idea in hopes that y’all who are smarter than me can work it out with me.

If you haven’t yet, check out manus.im. They’ve packaged into a consumer product the best practices of multi-agent orchestration, it’s really impressive.

I’ve got a refer a friend code, good for 500 credits (whatever that means). If you sign up, I also get 500 credits.

The big unlock over the last 6 months has been using 1) subagents to create an assembly line of specialists who divide and conquer a task, overseen by an orchestrator agent, 2) having the agent create a plan of attack before jumping into execution, and 3) using .md files (text files formatted in Markdown) to capture context for use by other agents. This solved for the problem of AI quality tanking because there was too much information being shoved into the prompt and context window.

They’ve also equipped Manus with the ability to open up a browser window and use it (scroll and click), run web searches, and write and deploy code. So if you need it to, say, create you a slide deck, it’ll break down the problem into pieces, interview you about what you want to say, generate the content with formatting and layout, use google image search to find relevant and appropriate stock photos for your slides, generate the code that is the slide deck, and deploy it to a URL for you to view and share.

If you’re new to creating agentic products, definitely give manus.im a spin and peer into how it is executing under the hood (it shows its work at each step). The way it tackles problems is worth emulating.

1  Which reminds me of how Facebook for years was an Aggregator wanting to be a Platform (remember Facebook apps?). It reeks of “it will be huge for us if X were true” product management thinking, the kind of thinking that forgets to start from the user and work backwards.

Their recent launches of Sora (as an AI video social media app) and Pulse (as a feed of ideas you didn’t ask for app) fit into the same theme — the future would certainly be better for OpenAI if these products were successful. But is it actually better for users?

Who knows, they might be right. But I’m not taking that bet.

Definitely check out Ben Thompson’s analysis and Casey Newton’s take.

The much more charitable analysis is they are taking a venture capital portfolio approach of making a bunch of bets on inventing the future, hoping that one is a home run amongst a sea of failed launches, and having so much cash to deploy that it’s a rounding error if most fail. But the real question is, what’s the opportunity cost? It’s a land grab out there right now.

2  WeChat, in China, is probably the biggest superapp. 1.3 billion users, using it for messaging, e-commerce, digital wallet that works with physical kiosks, and services like booking doctors appointments.