BYOCompute, Pope Destroys False Idol, Ben Affleck's White Pill | Joe Gebbia, Molly Fowler, Tom Dotan, Pim de Witte, Greg Castle

John Coogan (0:00)

You're watching TBPN.

Jordi Hays (0:02)

Today is Friday, October 2nd, 2026

We are live from the TBPN Ultradome, the Temple of Technology, the Fortress of Finance, the Capital of Capital. How many days till Christmas?

Tom Dotan (0:14)

83 days.

Pim de Witte (0:15)

But you already knew that. We don't even have to tell you at this point.

Jordi Hays (0:18)

The road to Christmas is going strong.

Pim de Witte (0:20)

That's right.

Jordi Hays (0:21)

Get ready, lock in. Also, sign up for RAMP. Time is money, save both. Easy use corporate cards, bill pay, accounting, and a whole lot more.

Greg Castle (0:29)

All in one place.

Jordi Hays (0:35)

Oh, we're getting Christmassy up in here. OK. Here we go. Very Christmassy. It's Q4. We're in the correct season now. So we can get pretty Christmassy.

Pim de Witte (0:45)

That is right.

Jordi Hays (0:47)

It's pretty tricky to stay in the little reef there.

Pim de Witte (0:49)

Enough about Christmas, John. Let's get into the news.

Jordi Hays (0:52)

Let's talk about artificial intelligence. Sounds great. Yeah, so make it less Christmassy. It's too Christmassy. It's 83 days away.

This is no more than 20 days away before we get that Christmas.

Pim de Witte (1:07)

I'll just leave this on.

Jordi Hays (1:08)

Okay. Just one level of Christmas.

Pim de Witte (1:11)

That's right.

Jordi Hays (1:12)

October 7th, October 2nd.

Yeah, so there was this funny moment in a Dev Day this week where everyone's focused on the war of dots versus muse versus instinct. And there were a couple of times where you were like, I want to talk about sign in with ChatGPT, which is the most boring and minute feature and very much not in the drama war for the personal agent. Who's going to do your shopping? Who's going to book your accommodation?

Pim de Witte (1:41)

I thought it was cool for independent developer, small app builders.

Jordi Hays (1:44)

It's actually probably more immediately relevant to the community that was there, the developers, right?

Pim de Witte (1:51)

Yeah. And you remember when, flashback to two years ago, people were creating these magical new product experiences with the best models at the time, but they were, in hindsight, pretty rudimentary. And they would build something magical, scale up really quickly, and then be like, oh, I just got stuck with like a $30,000 API bill.

Jordi Hays (2:13)

Yep.

Pim de Witte (2:13)

And so this is the solve for that, because it allows users to bring their own...

Jordi Hays (2:18)

Compute.

Pim de Witte (2:18)

Compute.

Jordi Hays (2:18)

BYOC.

Pim de Witte (2:20)

BYOC.

Jordi Hays (2:20)

Everyone knows BYOB. Now we're in the era of BYOC.

And yeah, like, you know, most companies, they could usually figure out a way to say, hey, like, this service is going to use a ton of compute, so you got to jump on the pro plan immediately. You got to set up a consumption, a credit card relationship, API credit, something like that, metered billing. But it was just an extra step, an extra point of friction of you saying like, okay, I already paid 10 bucks a month for this thing, now I got to actually pay for inference, and this gives like a different off take. So yeah, Peter Yang was feeling the same way that you were. He said, I'm excited about DOTS, but even more excited about sign in with ChatGPT feature. Letting users use the ChatGPT plan in your website or app makes a bunch of Indy apps a lot more viable.

Example, I have a live language learning app that I wasn't planning to launch to users due to API costs, but letting users sign in with ChatGPT makes this viable. Hope they roll it out to non-enterprise folks soon. Vignesh, who's actually running the project, said, I'm going to try and answer a bunch of questions that people have raised about signing with ChatGPT in the next couple of days. Before that, I want to let you know why they shipped it. He said, and so you can go read that thread, the bring your own subscription stuff, it sticks out because it feels like it's potentially like the beginning of a new economic structure. Like, is this an app store where there will be a margin that accrues to OpenAI or is this more just like better for discovery? Will agents reach for these because the sign-in is available? How much more smooth will this get? But at a high level, I think you described it well, a user can sign in to an app or service.

The notion example is a good one because it's very intuitive, people already pay for Notion. You can imagine all the AI features that are there and people were probably on day one of ChatGPT launching, copying from Notion into ChatGPT, copying back into Notion, writing bindings. But if Notion wants to offer a whole bunch of inference opportunities and agentic features inside, then they gotta go to other clients and say, hey, you're paying 10, 20 bucks a month for your seat. We also need a bunch of money for inference and this offsets that. And so the way it works is a user signs in to an app like Notion, then authorizes the app to run the OpenAI model request against their plan without managing an API key. On Notion, this means drawing from the ChatGPT allowance instead of Notion credits. And it makes a lot of sense for bigger companies that have inference bills, as you mentioned. And they've all had to grapple very quickly with product tokenomics, which is just a new economic muscle and price elasticity curve. They need to ride effectively. But the big thing is that it should, over time, allow leaner teams to offer inference hungry features with way less friction. So it's worth clarifying that the feature only brings OpenAI tokens along for the ride. It's not OpenAI dollars, which is, I think, a way where this could go. But the good example is, let's say you have some sort of AI book publisher where you're doing a bunch of token inference to generate the text of the book. Like you say, hey, I want to go take this general story, turn it into a book, and then print it and send me the book, right? There's two costs there. One's inference, one is-