[BidClub_]
Sharp Tech · · 16 min

(Preview) What Memory Can Do for ChatGPT, OpenAI's Future and Google's Circles Past, What College Football Can Teach Us About Tech

Andrew SharpBen Thompson

Podcast
TL;DR
  • Ben’s core call is that OpenAI’s API business consumes scarce capacity and demands backward compatibility, stability, and an enterprise culture that conflict with building a groundbreaking consumer product. His ideal split puts API maintenance and channel work at Microsoft, accepting its margin so OpenAI can improve the odds of its consumer “moonshot.”
  • Ben finds the memory vision compelling because “sign in with ChatGPT” could carry a user’s AI persona and circle into purpose-built apps, not merely authenticate identity. A Paprika-like recipe app illustrates why: text may be “the natural interface,” but the companion becomes more useful when expressed through interfaces suited to each task.
  • This concession does not rescue the existing API strategy: Ben calls the portable-companion concept compelling while describing today’s business as “backwards justification” for an API OpenAI started and now finds difficult to kill. He sees the companion concept as a different platform business from random companies using the API; a diminished Microsoft relationship may also prevent a seamless handoff.
  • Apple and Google have the cleaner integration path because their phone platforms can see screens, access device memory, and insert AI across applications. Ben says OpenAI is already large enough to make a “sign-in with ChatGPT button” attractive across the web.
  • Google Circles demonstrates that sound product logic is insufficient without timing, user cognizance, culture, and an installed social network. Ben’s broader lesson is that “success is about more than ideas”: users now manufacture separate circles through group chats, networks, and personas, but Google+ lacked the network when the feature launched.
  • ChatGPT’s 800 million users versus Anthropic’s 30 million make distribution—not technical capability—the decisive strategic advantage. Ben argues that only OpenAI, Apple, or Google can plausibly build the ubiquitous companion, and calls OpenAI’s “right place at the right time” an opportunity to become “the biggest company since Facebook.”
Digest · the substance, structured for research

1. OpenAI should let Microsoft carry the API burden

  • Ben’s overarching critique: OpenAI’s API is “a distraction and a waste of resources.” Enterprise platforms must preserve standards and backward compatibility, while API customers consume capacity that OpenAI itself describes as constrained.

  • His preferred division gives Microsoft the channel and “all the dirty work” of maintenance. Andrew notes that this does not forsake consumer upside; Ben replies, “No, it enhances it,” because platform stewardship and groundbreaking product development demand fundamentally different cultures.

2. Memory turns login into a portable AI companion

  • What Ben found compelling in Sam’s interview was the memory/companion concept, not a reversal of his API critique: a ChatGPT login might carry “your AI, your circle” into another app rather than merely verify identity. Confining that companion to OpenAI’s interface would constrain a persona meant to follow users across contexts.

  • Ben also argues that the chatbot interface is underrated and text is the natural interface, but it is not the best interface for every task. His concrete example is recipes: a user could open Paprika and find the companion manifested in a purpose-built kitchen experience.

  • Andrew proposes that APIs preserve muscles for this next layer; Ben rejects that as “backwards justification” for a business OpenAI began and now finds difficult to kill. He treats the portable-companion idea as a different platform business from random companies using the API. A diminished Microsoft relationship may also prevent a clean handoff, though portable memory remains a real update to his critique.

3. Device owners begin with a structural integration advantage

  • Apple and Google can see what is on-screen, access device memory, and insert AI into conversations. ChatGPT must instead secure partnerships and persuade services to add its login.

  • Ben says OpenAI is large enough to ask services to add a “sign-in with ChatGPT button.” Andrew’s comparison: carrying an AI companion may be potentially far more valuable than dragging a Facebook profile—and its college photos—around the internet.

4. Google Circles failed because timing and path dependence dominate ideas

  • Ben calls Google Circles “a great idea,” but it lacked user cognizance, a social network, and the right company culture. Ideally, Google+ would have sorted people into circles, but sorting a life was useless when the relevant people were not there.

  • His systems lesson is broader: complex systems emerge from many interacting conditions and resist perfect top-down design. Today users approximate circles through group chats, separate networks, and different personas, though the absence of an installed network remains an obstacle.

  • The super-app comparison makes path dependence concrete: Americans moved from PCs and fragmented communications to smartphones, producing “a screen full of apps.” In China, most people’s first computer was their smartphone, letting WeChat become the natural place to launch services; Zuckerberg’s Facebook-phone vision arrived too late. Had the Facebook phone existed a decade earlier, Ben says, it might have worked.

5. ChatGPT’s installed base creates an option rivals cannot copy

  • Ben says there is “nothing technologically” preventing Anthropic from building the same companion, but contrasts its 30 million users with ChatGPT’s 800 million. Andrew crystallizes the constraint: “The hard part is getting to a billion users.”

  • Apple and Google remain plausible alternatives, though each is bounded by its phone platform. Ben’s categorical conclusion is that no other foundation model company can take ChatGPT’s place; OpenAI’s rare advantage is being “the right place at the right time.”

Andrew Sharp

Hello, and welcome to a free preview of Sharp Tech.

Matthew says, “Ben and Andrew loved the latest episode on Facebook’s 3 eras, and one thread jumped out to me. The way OpenAI is talking about platform ambitions publicly and in Sam’s recent interview with Ben reminds me a lot of Facebook circa 2010 to 2012. Back then, Zuckerberg talked about building an OS layer, SDKs, third-party developers, the whole thing. It felt like a big bet on becoming the platform everyone else would build on, and I’m starting to hear echoes of that same narrative in how Sam Altman frames OpenAI’s role today.

“Some of this might be structural. Venture-backed companies seem to get nudged toward platform strategies, and it’s a recurring ambition in tech, but one that often runs into the same walls. Do you see the same parallels, or does OpenAI’s positioning feel fundamentally different this time? Ben, what do you think?”

1. OpenAI Must Pick A Lane

Ben Thompson

I thought this was a really interesting email that was very thought-provoking. Sam Altman’s discussion—just for context, I’m critical of the OpenAI API broadly speaking, in that I think it’s a distraction. I think what it takes to be a good enterprise company is that you have to maintain old standards and backwards compatibility.

And, oh, by the way, OpenAI always talks about capacity constraints. Well, some of their capacity is being used by API customers, or a lot of their capacity is being used by API customers. I just think there’s a real fundamental disconnect in pursuing a consumer company and trying to be an API company. I think those are 2 very, very different jobs, and I think it’s—

Andrew Sharp

Mm-hmm.

Ben Thompson

—it’s a distraction and a waste of resources for a consumer company to be trying to be an API company as well. This is where, in an ideal world, the Microsoft relationship should have worked out, where Microsoft is the API provider. That’s what they do. That’s what they’re good at. And OpenAI is pushing forward.

It goes back to the article I wrote, “The Shame of Mobile,” like a decade ago. I think I was writing it in the context of the iPad. The shame of this era is that the price to become a platform is that you have to build a great product.

But all the skills that go into building a great product make you kind of a crappy platform provider for developers, because the whole point of being a platform is that you’re letting them shine and make great products, and you’re just being stable. You’re not competing with them. You’re maximizing what you can do for developers.

The shame of it is that Microsoft is a much better platform company, but they suck at product, so they will not be a platform company in this era. Apple is a great product company, and they suck at being a platform, but we’re stuck with them.

Andrew Sharp

Mm-hmm.

Ben Thompson

In an ideal world, OpenAI would have internalized this and said, “Okay, if we’re really going to lean into being the best possible product company we can be, by definition, it’s actually good that we be a bad platform company, because the skills and capabilities necessary are inversely correlated.”

Microsoft is great at being a platform company. We can let them do it—let them do all the dirty work, all the maintenance, all those sorts of things. They have the channel. We can basically monetize our technologies through them, not to the greatest extent theoretically possible because they have to take a margin too, but we can free ourselves of this burden and push forward and be the best possible consumer company we should be.

That’s my thesis.

Andrew Sharp

Mm-hmm.

Ben Thompson

And I still stand behind it.

Andrew Sharp

And it’s not like that thesis forsakes their upside on the consumer side.

Ben Thompson

No, it enhances it.

Andrew Sharp

And that’s the point.

Ben Thompson

That’s the point.

Andrew Sharp

Right.

Ben Thompson

You can do more and do better. There are just fundamental skills—

Andrew Sharp

They’re still taking the moonshot. They’re just taking a smarter moonshot by offloading the really annoying, boring responsibilities—

Ben Thompson

They’re increasing the likelihood—

Andrew Sharp

—to Microsoft.

Ben Thompson

—that they achieve the moonshot because they’re putting all their wood behind one arrow. And it’s not just that it’s different skills. It’s different cultures. The culture of—

Andrew Sharp

Mm-hmm.

Ben Thompson

—a platform API provider is fundamentally—

Andrew Sharp

Yeah.

Ben Thompson

—different from that of a groundbreaking—

Andrew Sharp

Well said.

Ben Thompson

—product maker. To the extent they need to be a groundbreaking product maker is the extent to which they are diluting their effectiveness and their long-term capability by trying to be a platform. So that’s my overarching critique.

Andrew Sharp

Okay. But I sense a “however” coming here in response to Matthew’s email. Am I right?

2. The AI Companion Opportunity

Ben Thompson

What I thought was compelling about what Sam said to me in that interview was this concept that there are a lot of product needs where they can’t fulfill all of them. This memory bit speaks to it: Could I take my memory? Could I take my AI persona and bring that into other apps where you sign in? Signing in with ChatGPT doesn’t just mean it’s an identity login. It means you’re bringing your AI, your circle—

Andrew Sharp

Mm-hmm.

Ben Thompson

—into other experiences.

Take this recipe example. At the end of the day, again, I think the chatbot interface is underrated. I think text is the natural interface. I wrote “Messaging Is Mobile’s Killer App” in 2012 or 2013, literally the day before Facebook acquired WhatsApp—one of my luckiest timings of all time—where I basically laid out the rationale for this acquisition before it came out.

But at the end of the day, it’s not the best interface for a recipe app. I want a recipe app that I can see on my phone very quickly in the kitchen, or on a screen, or whatever it might be, so I can interact with it. So wouldn’t it be better for Matthew if he could sign in to Paprika, a great recipe app, and it’s just there and manifested in a way that’s appropriate to the use case at hand?

I think that’s a compelling vision. I will just note that I don’t think Sam actually answered my concerns about the platform. That’s a different platform business from random companies using our API. Now, maybe—

Andrew Sharp

Yeah.

Ben Thompson

—the argument is that by using our API, we can layer in the login on top and it all sort of ties together. It’s like a—

Andrew Sharp

So they’re maintaining their muscles in order to take the next step once the—

Ben Thompson

No, I think what it is is backwards justification for the fact that they started out as an API business, and it’s very difficult to kill. That’s the whole problem with API businesses. Once they’re out there, they’re out there.

Unfortunately, I think the relationship with Microsoft has probably diminished to the point where they can’t just seamlessly hand it off, or whatever it might be. But that vision of carrying your companion with you is really compelling. It actually fits with what I’m talking about.

Part of the idea of the AI being with you is that limiting it to 1 user interface, which is whatever OpenAI provides, is a constraint on that vision, because my circles do not all live within one app. It would be nice—

Andrew Sharp

Mm-hmm.

Ben Thompson

—to be able to bring that with me.

This also speaks to the Apple and Google opportunity, which is that the easiest way to have an AI with you is to control the underlying platform where you interact with all these devices. Google and Apple can see everything on screen. They have access to memory on your device. They can insert themselves into any and all conversations in a way that ChatGPT needs to get all these partnerships and sign people up to be able to bring it in, which is much more challenging.

No, I think that part of the vision is compelling. That is an update on my critique. I still stand by my critique, but I do think that idea of carrying your AI with you to different experiences is a good one. And they’re the only ones that can do this. They’re the only ones that are large enough—

Andrew Sharp

Yeah.

Ben Thompson

—to have enough users where they could go to everyone and say, “You should add a Sign in with ChatGPT button.”

Andrew Sharp

And it’s all potentially a lot more valuable and compelling than being able to take your Facebook profile with you anywhere you go on the internet. In that respect, OpenAI’s positioning may actually be—

Ben Thompson

Right.

Andrew Sharp

—a little bit different this time.

Ben Thompson

That goes back to our initial critique, which is: Why do I want my friends and family with me in this app, right?

Andrew Sharp

I don’t want to sign in—

Ben Thompson

That’s right.

Andrew Sharp

—with my Facebook account back in 2010, when it was full of drunk photos from my college years.

But good email, Matthew. And with that, we have 1 other question on ChatGPT. John says, “Regarding Ben’s take on ChatGPT memory, didn’t Google try to do this with Google Circles? The colored ovals that comprise the different facets of Ben—that was what Google Circles was trying to do as a social network. You could present different aspects of yourself to different circles.”

I thought it was a cool idea, but it didn't seem to catch on. Was it because the tech just wasn't ready yet, or because we weren't yet so ubiquitously online that we all had to adopt different personas for different forums? Today, for instance, one person's posts to Facebook are different than their posts to LinkedIn and different than posts to a group chat with a bunch of college fraternity brothers.

Ben, do you have thoughts for John? A little Google Circles history here.

3. Google Circles Needed A Network

Ben Thompson

Yeah, no, this came up at the time I wrote this article. A lot of people responded to this. Because, yes, that is the idea of Google Circles, and it was a great idea.

Andrew Sharp

Mm-hmm.

Ben Thompson

The problem is that success is about more than ideas. I did a follow-up daily update to Social Networking 2.0, which I just clicked through here, and my section on this is called “Circles and Timing.” Timing matters. At that moment in time, number 1, there wasn't the cognizance of this, which John sort of admitted.

There's also the fact that Google Circles had no social network. You needed that; that was the prerequisite. Ideally, I could sort everyone in my life into these circles, and Google+ would handle that. But none of the people in my life are on Google Circles. That's a real obstacle to this.

Yes, there are a lot of examples of tech history like this, and tech is not enough, and Google’s not a social networking company. They don’t have the right culture. There are so many pieces that go into this being the case. There’s a broader point about complex systems versus designed systems.

We talked about this in the context of manufacturing and tariffs and all these sorts of bits and pieces. I’m very skeptical of attempts to design perfect systems from the top down.

Andrew Sharp

Mm-hmm.

Ben Thompson

Complex systems are emergent because there are so many things that go into them and make them work that matter. In this case, it makes sense. If Google Circles came along today, a lot of people have manufactured this using group chats, using different social networks, or, in my case, even using different personas.

Something like that would make more sense today, but then you still have things like, well, no one's actually on Google Circles, right? And this is where the—

Andrew Sharp

Right.

Ben Thompson

—the question of the opportunity comes in, and everyone goes back to why everyone wants to be a platform. The lower down in the stack you are, the more you could theoretically manifest and show these sorts of things.

4. America Missed The Super App

But it's kind of like the old debate: Why aren't there super apps in America when they're in China and WeChat is the app for everything or whatever? It's like, no, that's never going to happen because of the nature of how this stuff came to market in the US.

You started out; most people had PCs. They went from PCs to smartphones, and there was already this real bifurcation of different communications channels and stuff. Whereas in China, most people's first computer was their smartphone, and it was much more viable to create the everything app because it was the first thing people got, and that was a natural place to build it out.

You didn't have that path dependency—the path dependency really matters. That path was closed to the US. That's why we ended up—

Andrew Sharp

Mm-hmm.

Ben Thompson

—with a screen full of apps. Again, I think this is another area where Mark Zuckerberg really got this wrong, going on and on: “No one wants a bunch of apps. They want to be centered around the people.” No, actually, that’s wrong. They do want a bunch of apps because that’s where we were at.

Go back to the Facebook phone in 2012 or 2013 or whatever it was. It's like, no, it's too late. If the Facebook phone existed a decade ago, maybe, but it didn't. Whereas in China—

Andrew Sharp

Yeah.

Ben Thompson

—that was how it emerged because people were getting their first smartphones in 2011, 2012, 2013, whenever, and WeChat was there, and that was the killer app right away. There, it was the opposite. It was easier to launch something in WeChat, which everyone had, instead of expecting people to go get apps.

Andrew Sharp

Mm-hmm.

5. ChatGPT Has The Consumer Lead

Ben Thompson

Path dependency matters hugely, and it matters to this discussion and this point John is raising. Google Circles was a great idea. Great ideas are one piece of many. And it also—

Andrew Sharp

Right.

Ben Thompson

—just to circle this back to ChatGPT, being where they are, when they were, and getting all that consumer mindshare opens up all these possibilities to them that are not open to anyone else.

Andrew Sharp

Mm-hmm.

Ben Thompson

If ChatGPT does not become this companion with you that goes everywhere and is tied into all these apps, it's not going to be any other foundation model company. It's not going to be Anthropic. It's not going to be Grok. The only other possibility is that it's one of the phone companies: Apple or Google. But even there, they're constrained to an extent by virtue of just being Android or just an iPhone.

Andrew Sharp

Yeah.

Ben Thompson

That's why I'm so strident about OpenAI going for this and why I say this has the potential to be the biggest company since Facebook, because they're in the right place at the right time, and those are actually the rarest possible qualities.

6. Scale Is The Real Moat

Andrew Sharp

And when you say it's just not possible for a company like Anthropic, just to tease that out, it's because they lack the scale of users to make it a meaningful—

Ben Thompson

Right. There’s nothing—

Andrew Sharp

—selling point.

Ben Thompson

—technologically preventing it from being Anthropic.

Andrew Sharp

Right.

Ben Thompson

The problem is they have—

Andrew Sharp

Yeah.

Ben Thompson

—30 million users, and ChatGPT has 800 million users. Just because ChatGPT was first—

Andrew Sharp

Yeah.

Ben Thompson

—and they captured that.

Andrew Sharp

The hard part is getting to 1 billion users. And Google Circles—maybe Google Circles could have succeeded if they had gone back to 2003 and launched a social network that took the entire world by storm, but—

Ben Thompson

If they owned your identity, like I'm talking about, and then could help you parcel out your identity, yes, that would be compelling. They didn't own identity. Facebook did. It was already too late.

(Preview) What Memory Can Do for ChatGPT, OpenAI's Future and Google's Circles Past, What College Football Can Teach Us About Tech | BidClub