Your Name - Personal Blog

What is the value of the internet?

I think about the internet like one giant, realtime database. In the simplest sense, all it really is, is a set of servers that hold data about everything. And while I think that’s important, I also think it’s something that will change very soon.

Over 51% of internet activity is botted. It’s inevitable that with new visual and text based content generation, it’s going to become less and less human and more just bots scrambling for attention and data. It also is clear to me that humans will shift less and less towards using the internet, and more towards having their own digital representatives interacting with them on the web.

Entire industries are piggybacked on this idea of scrambling for data. Every sales company builds a proprietary data moat, marketing teams pay thousands of dollars for competitor or user analysis, and recruiting is purely a game of sourcing and ranking.

But data INTERACTIVITY is a different question altogether. Why would anyone ever want to, say, ever again go to Google Maps, search for a dinner spot, find one with a nearby subway stop, and then finally book a reservation with resy? Agents can do that today in one shot, if given the ability and permissions to integrate with these different entities.

What's needed now?

I think I say or hear the word agents every single day now. Hard not to if you’re in startups. But this also means there’s a ton of noise in the ether regarding what agents are needed, which ones aren’t, etc. Perception and attention seem to be the major differentiators today, and I learned this firsthand at Oleve.

Inherently, the mastery of how you handle what users want and what they don’t want is essential to figuring out how to build a great agent (or just product in general), and becoming an excellent AI engineer means understanding your users. Websites do that, in a sense. They appeal to more human tendencies, hence having excellent HUMAN user experience is what people focus on when building landing pages.

After working on AI agents for consumers, sales, engineering, and recruiting, I can tell you firsthand the pattern is the exact same. No one cares about software.

But there's a massive opportunity now for building agent centric experiences.

With the release of the OpenAI Apps SDK, they're confirming they want to head down the same track I believe we're all moving towards, which is a monopoly over the agentic internet. One where companies won't have to or want to optimize for the human eye, but for being called as a tool.

The OpenAI Apps SDK - Towards a new internet?

If you have some MCP server, you can connect it to ChatGPT and serve your product as a tool call to the millions of ChatGPT users. I think of this as a circumnavigation around GEO to directly embed your product in a generative engine, rather than biasing it towards mentioning a link to your website. It also opens up an entirely new primitive of how agents can interact with products. You don’t need some browser use agent with special privileges, you can just build the needed tools to an agent to execute interactions on your app directly, without having to worry about much of anything else.

I think there's a few big winners I can see here:

1. Agentic native interface apps.

a) Some social app that works only between ChatGPT users will dominate, I can easily see a connection based prosumer app release and grab up a massive share of the 100M+ DAU ChatGPT serves. If anyone ever hops on gpt and asks for a connection / warm intro to someone, this would be solid gold. Bonus points to use simulated interactions with other users to check best-fit.

b) An educational UI layer on top of GPT. Although I think Khan Academy is soon releasing an integration, I'm very interested to see how education changes in ChatGPT.

2. Observability

a) The most important thing to observe is the tool calls that gpt makes with your product, and the user queries that lead to said calls. But I don't even really know a) how much openai plans to expose to developers, or b) whether tweaking the tool definitions is an engineering, growth, or product problem.

b) My gut says all three, with the likely outcome being the target user here. I would guess there's going to by many different solutions for different types of people.

3. Essential Infrastructure

a) I struggle with the differentiation of this a little bit. While I don't think it's as simple as just claiming to be the 'Vercel for ChatGPT apps', the appeal of building that sort of product is extremely real. Although, incumbents might blitzkrieg-style dominate all the functionality here, so I'm not entirely sure whether or not there's real viability to building something here.

b) If you want to stand out here, try to first understand what it means to truly be an agent facing product. My guess is when building some of these apps that would work as a gpt tool, but not with a website or app, there will be some unique infrastructure requirements for specialized UI components, models, communication protocols, and more.

4. Physical world interactions

a) Some companies already have this covered (Uber will be launching on apps sdk, so will DoorDash), but these companies have spent a decade optimizing for human experience, not agent experience.

b) I'd really love to see someone try to take on a "background chores handling app", that basically allows you to input all the different kinds of busywork you have in the day, then orchestrates food delivery, dry cleaning, having a robot do your dishes, etc. etc. to perfectly run every non-essential part of your life in one perfect interaction layer.

Only one thing has ever helped me confirm/deny theories about anything; building and yap. For the next few months I’ll be building several apps, tooling, and infrastructure around this layer, both for myself and some select design partners. As a AI systems developer by trade, obviously the infrastructure and observability appeals to me the most, but I definitely am not closing any doors.

If you’re interested in building the next wave of consumer tech, or want to join me on this journey, reach out.

Thanks for reading this far.

-Abhi