Onpode
Cover art for OpenAI's personal agent, Meta's protocol, and a new standards war are reshaping AI workflows

OpenAI's personal agent, Meta's protocol, and a new standards war are reshaping AI workflows

October 7, 2026 · 8 min

Ryan Castillo & Jordan Hale

OpenAI launched Dots, an always-on autonomous agent connected to 4,000+ apps, at DevDay on September 29 — with no public API. Five days later, Meta and Sierra announced the Personal Agent Protocol with Stripe, Shopify, and Walmart as founding partners but no published spec. Developers have nothing stable to build against.

On September 29, 2026, OpenAI introduced Dots at DevDay in San Francisco — a new class of always-on autonomous AI agents built into ChatGPT. Each Dot runs on its own cloud computer and browser, pursues user-defined goals continuously in the background, and connects to over 4,000 apps via OpenAI's plugin ecosystem.

0:008:22
Get the next episode on Technology →

Follow it free — new episodes land in your feed.

Or make your own — any topic, in minutes

More Onpode episodes on Technology →

About this episode

OpenAI held DevDay on September 29th and introduced Dots, an always-on autonomous agent designed to handle tasks continuously — no prompt required, no app to open. The framing was bold. The reality on launch day was a preview for premium subscribers and no public API for developers to build against. Five days later, Meta and Sierra announced the Personal Agent Protocol, with Stripe, Shopify, and Walmart signing on as founding partners before the technical spec had even been published. OpenAI and Anthropic — the two labs whose models underpin most of the agents PAP is meant to govern — weren't in the room. This episode traces the gap between announcement and usable reality, and what it means for the people caught in the middle. Meta's distribution moat is not a technical argument: 3.6 billion daily users across WhatsApp, Instagram, and Facebook is a different category of leverage than a premium subscriber rollout. Walmart understood that, which is why it showed up at both launches. The episode also looks at where Sierra fits — $950 million raised in May 2026, $15 billion valuation, over $150 million ARR — and why that capital was already betting on this coalition before any standard was announced publicly. The concrete question underneath all of it: a developer building checkout automation right now has three incompatible paths, real rebuild costs if she picks wrong, and no stable ground under any of them. The race is already happening. Nobody is waiting for the rulebook to ship.

Frequently asked

What is OpenAI Dots and how is it different from ChatGPT?

OpenAI Dots is a persistent, always-on autonomous agent that runs continuously on OpenAI's cloud computers — monitoring email, Slack, and 4,000+ connected apps without waiting for user input. Unlike ChatGPT, which responds only when queried, Dots pursues goals over time. It launched in preview for ChatGPT Pro and Business Premium users with no public API.

What is Meta's Personal Agent Protocol (PAP) and who are the founding partners?

Meta's Personal Agent Protocol (PAP) is an open standard for AI agent interoperability announced October 6, five days after OpenAI DevDay. Founding partners include Stripe, Shopify, Walmart, Genesys, Rocket, NiCE, and Decagon. OpenAI and Anthropic were absent. The full technical spec was not published at announcement and was expected later in October.

Why did Walmart join both Meta's Muse and the Personal Agent Protocol?

Walmart announced a Meta Muse partnership at Meta Connect in September, then appeared as a PAP founding partner in October. The dual commitment reflects a hedge strategy: enterprise partners are not evaluating competing specs on technical merit — they are following user reach, and Meta Muse touches 3.6 billion daily users across WhatsApp, Instagram, and Facebook.

What is the Agentic AI Foundation and has it produced any usable standards?

The Agentic AI Foundation is a Linux Foundation coalition with OpenAI, Anthropic, Google, Microsoft, and AWS as founding members, formed in December 2025. Despite having the broadest industry representation of any agent standards body, it had produced no standard with real developer traction as of the PAP announcement.

Should developers build on OpenAI Dots, Meta PAP, or Anthropic MCP right now?

As of the PAP announcement, no option offers stable ground: Dots has no public API, PAP has no published spec, and Anthropic's Model Context Protocol works but only cleanly inside Anthropic's partner ecosystem. Developers choosing a stack are betting on momentum over the next eight months, not on a finalized technical standard.

Grounded in 6 sources
Microsoft overhauls Copilot to take on AI rivals like Anthropic - CNBC ↗ · cnbc.com
https://techcrunch.com/2026/10/06/the-next-hurdle-for-ai-agents-getting-websites-to-let-them-in/ ↗ · techcrunch.com
AI Agent Interoperability: What MCP and A2A Mean for Enterprise Governance | Airia ↗ · airia.com
Meta's Personal Agent Protocol Signs Walmart Before It Has a Spec | THE D*AI*LY BRIEF ↗ · beri.net
Genesys Joins Sierra, Meta on Open Standard for Personal AI Agents ↗ · cmswire.com
Interoperability Is Key To Unlocking Agentic AI’s Future ↗ · forrester.com
Read transcript

Ryan Castillo: Hey — rough week for anyone trying to make sense of the agent space, or just me?

Jordan Hale: Just you, clearly — no, I mean, it's been a lot. There's this moment I keep replaying where OpenAI is on stage in San Francisco calling Dots an 'iPhone moment,' and I'm thinking, they shipped zero public API. Zero. That's the thing that's stuck with me.

Ryan Castillo: Before we get to what's missing — what did they actually ship? Walk me through it.

Jordan Hale: So DevDay, September 29th, San Francisco — OpenAI introduces Dots. And the core idea, stripped of all the language, is: imagine hiring an assistant who never clocks out. Never waits for you to show up and ask. Just quietly handles your Slack, your email, whatever you've pointed it at — while you're asleep, while you're in meetings, continuously. That's the concept. An always-on autonomous agent, running on its own cloud computers, connected to four thousand plus apps, reachable through Slack and Teams and ChatGPT. And under the hood it's GPT-6 Astra.

Ryan Castillo: Wait, so it's literally running when I'm not using it — like it doesn't need me to open anything?

Jordan Hale: That's the whole break from chatbot logic. A chatbot is passive — you query it, it responds, done. Dots is persistent. It holds context, it pursues goals over time, it's not sitting idle waiting. That distinction is — I mean, that's actually new. The rollout is ChatGPT Pro and Business Premium users, eligible markets, preview only for now.

Ryan Castillo: Preview only, no public API — so the number that matters here is zero. Zero developer access on launch day.

Jordan Hale: Which is wild given the framing, you know? You don't call something an iPhone moment and then tell developers to wait.

Ryan Castillo: And then five days later — five days — Meta and Sierra announce PAP. October 6th. That timeline is not an accident.

Jordan Hale: Wait, Bret Taylor announced this? Like, personally, on X?

Ryan Castillo: Two thousand seven hundred likes. Two hundred eighty-two thousand views. He framed it as 'open for anyone to implement.' Bret Taylor is Sierra's co-founder. He is also, simultaneously, OpenAI's board chair.

Jordan Hale: Okay that's — I mean, that's a genuinely strange position to be in. You know, you're announcing a competing protocol while sitting on the board of the company whose product just launched five days before with no API? Like, did OpenAI just... not get called?

Ryan Castillo: That's the question nobody's answered. Founding partners on October 6th: Stripe, Shopify, Walmart, Genesys, Rocket, NiCE, Decagon. OpenAI — absent. Anthropic — absent.

Jordan Hale: Not absent like they declined to comment. Absent as in they're not there at all.

Ryan Castillo: Right — but the part that doesn't fit is: those two companies power the vast majority of the agents this standard is supposed to govern. So what does that mean for the standard's actual reach on day one?

Jordan Hale: And — wait, actually — the full technical spec wasn't even published at announcement. Like, the spec that everyone's supposed to implement? Expected later in October. So you've got a standard with no spec and no buy-in from the two biggest model labs. That's not a protocol yet, you know? That's a press release with good distribution.

Ryan Castillo: Which makes the iPhone comparison even more corrosive. Dots: no public API. PAP: no published spec. A developer sitting down right now to build agent integrations has exactly zero stable ground to build on.

Jordan Hale: And that's the thing — the zero stable ground problem, it sounds like a technical issue, but it's actually a distribution issue wearing a technical costume. Like, forget the spec for a second. Meta Muse already touches 3.6 billion daily users. Not subscribers. Not premium tier. Daily users, across WhatsApp, Instagram, Facebook. OpenAI's Dots is rolling out to ChatGPT Pro. Those aren't even the same category of bet.

Ryan Castillo: 3.6 billion versus a premium subscriber base. That's not a gap — that's a different planet.

Jordan Hale: And Walmart saw it. Walmart shows up at Meta Connect in September, announces a Muse partnership — then shows up again at the PAP announcement in October as a founding partner. That's not loyalty, you know? That's a company following the eyeballs and hedging everything else.

Ryan Castillo: Which means enterprise partners aren't reading the spec. They're reading the MAU count.

Jordan Hale: Right — but the part that's undercovered is Sierra in all this. They raised $950 million in May 2026, $15 billion valuation, over $150 million ARR. That money landed before PAP existed publicly. So the capital was already betting on this coalition before anyone announced a standard.

Ryan Castillo: Platform wars don't get decided by the better spec. App Store, AWS versus Google Cloud — winner is whoever hits developer critical mass first. Meta's sitting on a distribution moat that took fifteen years to build. Dots can't outrun that.

Jordan Hale: Which reframes Dots entirely, I mean — actually, no, not reframes — clarifies it. Dots might not be losing to Muse. They might just be playing a completely different game. High-margin, power users, deep integrations. Muse runs on billions of devices at lower depth. Those could coexist.

Ryan Castillo: Could. Or one of them becomes the default and the other becomes a footnote. Someone has to pick today.

Jordan Hale: A single engineer at an early-stage company is making this call right now, staring at three incompatible paths with real rebuild costs on a wrong bet. We should get into what that decision actually looks like.

Ryan Castillo: Let's make her real. She's got a checkout automation integration to ship by end of month. She opens her laptop, she's got three tabs. Dots API — doesn't exist. PAP — Shopify is a founding partner, Stripe is a founding partner, those are exactly her vendors, but the technical spec isn't published yet. And MCP — Anthropic's Model Context Protocol exists, it works, but only cleanly inside Anthropic's partner ecosystem. She doesn't have six weeks to rebuild. Which tab does she close first?

Jordan Hale: Okay but — wait, that's not even a standards question at that point. That's just a timing question. She's not evaluating specs, she's evaluating who's shipping something she can actually touch.

Ryan Castillo: Timing is the question. That's my point. The open standard versus platform moat debate isn't philosophical for her — it's the thing she's staring at right now.

Jordan Hale: Right — but the part that's cruel is, like, the Agentic AI Foundation has literally every name on it. OpenAI, Anthropic, Google, Microsoft, AWS — all founding members, December 2025, under the Linux Foundation. That's the broadest coalition by far. And it has produced... no standard with any traction.

Ryan Castillo: So she's got the biggest tent and no usable output from it.

Jordan Hale: Exactly — and that's, I mean, that's actually the whole thing in miniature. The coalition with the most authority built nothing deployable. The coalition with a deployable standard — PAP, Stripe, Shopify as founding partners — hasn't published the spec. And MCP is technically real but, you know, it's narrow. She's not betting on a protocol. She's betting on who survives the next eight months with enough momentum that her integration doesn't become dead code.

Ryan Castillo: And that's asymmetric in a specific way — wrong bet isn't just wasted time. It's six weeks of rebuild at a moment when her competitors are also choosing. First mover on the right standard wins the integration slot.

Jordan Hale: No, and — wait, actually the image I can't shake is her just closing the Dots tab immediately. Not because Dots is worse, but because there's literally nothing to integrate against. You can't bet on an API that doesn't exist.

Ryan Castillo: Which means OpenAI's iPhone moment just handed PAP a free six-week head start with exactly the developer who'll decide which ecosystem gets embedded in the next generation of checkout flows. That's the unresolved tension nobody's comfortable sitting with.

Jordan Hale: You know what's strange — we started this whole thing with OpenAI on a stage calling Dots an iPhone moment, and... I keep thinking about the actual iPhone. You could walk into an Apple Store. You could hold it. Developers had a date. This is a moment that, like, hasn't happened yet. Someone is writing the first line of integration code right now, tonight, against one of these stacks — and they genuinely don't know if the thing they're building toward will still be the thing in eighteen months.

Ryan Castillo: Dots: no public API. PAP: no published spec. The race is already happening.

Jordan Hale: Yeah. And nobody's waiting for the rulebook to ship.

OpenAI's personal agent, Meta's protocol, and a new standards war are reshaping AI workflows · Onpode