Begun, the Aggregator Wars Have…

1. Key Themes

Aggregators are now fighting each other over the physical world, not just attention

Meta's new agent, Muse, exposes a structural tension: Amazon's core moat has shifted from digital attention to physical logistics, and that moat is now being tested by another aggregator rather than a traditional competitor.

"One of the best utilities Amazon provides is reliably delivering mundane things that are necessary and mundane. That means Muse is better if it can use Amazon, but that's a problem for Amazon because Muse doesn't see ads."

Physical-world investment as a durable AI-era moat

Amazon's blocking of Muse isn't just defensive friction — it reflects that Amazon's logistics infrastructure is a genuine competitive advantage in the agentic era that software-only players can't easily replicate.

"Amazon predictably blocked Muse, but there is room for a deal based on the reality that Amazon's physical world investments are an AI moat."

Proprietary data/platform control strategies can backfire when they degrade core user experience

GM's multi-year bet on ditching CarPlay for proprietary software (to capture data) ultimately collapsed under customer and channel pressure — a cautionary tale about prioritizing data capture over user experience.

"GM has announced it is re-integrating CarPlay after three years of customer and dealership backlash."

Frontier AI labs face structural pressure to keep pace, not slow down

There's an emerging discussion around whether leading AI labs can unilaterally throttle their own development given competitive dynamics — tying into broader questions about pacing, safety, and "overhangs."

"Pacing the frontier may be sincere, but it would also be strategically useful for the frontier labs to have time to reduce overhangs caused by model advancement."

2. Contrarian Perspectives

Betting against customer-experience-degrading platform lock-in strategies

Andrew Sharp initially thought GM's proprietary software strategy might not be a problem, but the consensus (or at least Ben Thompson's) view — that abandoning CarPlay was myopic — proved correct after three years, suggesting that data-capture strategies which sacrifice core UX face longer-term reversal risk even if they seem defensible short-term.

"I wasn't sure it would be a problem, whereas Ben was incensed at GM's myopia. Three years later now … Well, Ben was right."

3. Companies Identified

Amazon — E-commerce/logistics giant. Mentioned as blocking Meta's Muse agent to protect its ad-based business model, while its physical-world infrastructure is framed as an emerging AI-era moat. "Amazon blocked Meta's new agent... Amazon's physical world investments are an AI moat."
Meta — Tech/social media giant. Mentioned for launching Muse, an AI agent that needs Amazon's delivery infrastructure but threatens Amazon's ad model, and for needing Walmart's help to compete. "To bridge that moat Meta will need help from one specific company: Walmart."
Walmart — Retail giant. Mentioned as the key partner Meta needs to work around Amazon's blocking of Muse. "Meta needs Walmart to wait out Amazon."
Expedia — Travel booking platform. Mentioned in the context of trying to preserve its "middleware" position amid the rise of AI agents like Muse. "Expedia seeks to keep its middleware position."
Google — Search/tech giant. Mentioned as conspicuously absent/uncertain in the unfolding aggregator agent wars. "Meanwhile, where is Google?"
GM (General Motors) — Automaker. Case study in reversing a controversial proprietary software strategy after backlash, re-adopting Apple CarPlay. "GM has announced it is re-integrating CarPlay after three years of customer and dealership backlash."
Apple (CarPlay) — Mentioned implicitly as the platform GM abandoned and is now reinstating due to customer/dealer demand.
Colossus — Media/profile-writing outlet. Mentioned as home to Jeremy Stern's acclaimed profile writing, including a profile of Mark Zuckerberg. "Stern's recent profile of Mark Zuckerberg."
Anthropic / OpenAI — Frontier AI labs. Referenced regarding competitive dynamics preventing unilateral slowdown in AI development. "This week's Sharp Tech video is on why Anthropic or OpenAI can't unilaterally slow down."

4. People Identified

Ben Thompson — Author of Stratechery, host of Dithering and Sharp Tech. Mentioned as correctly predicting GM's CarPlay reversal and for conducting the interview with Jeremy Stern. "Ben was right, GM has announced it is re-integrating CarPlay."
Andrew Sharp — Stratechery/Sharp Tech contributor and writer. Mentioned as co-author of this roundup and for his shifting view on the GM/CarPlay debate. "I wasn't sure it would be a problem, whereas Ben was incensed at GM's myopia."
Jeremy Stern — Editor-in-Chief of Colossus. Mentioned as a top-tier profile writer interviewed by Ben Thompson, notable for his profile of Mark Zuckerberg and others. "Ben considers him 'the best profile writer working today.'"
John Gruber — Daring Fireball writer, co-host of Dithering with Ben Thompson. Mentioned as a recurring collaborator on the CarPlay/GM discussion.
Bill Bishop — Sinocism author, co-host of Sharp China. Mentioned in context of covering US-China relations and Xi Jinping's Washington visit.
Mark Zuckerberg — Meta CEO. Mentioned as the subject of Jeremy Stern's acclaimed Colossus profile, contrasted favorably against a fictionalized portrayal. "Stay for a behind-the-scenes discussion of Zuckerberg that appears to be much truer to life than that new Social Network sequel."

5. Operating Insights

Don't sacrifice core user experience for proprietary data capture — GM's multi-year experiment shows that even well-resourced companies with strong incentives (owning proprietary in-car data) can be forced into reversal when the underlying product experience suffers and customers/channel partners push back.
Platform gatekeepers will defend their moat against agentic disintermediation — Companies whose business model depends on attention/ads (Amazon) will act defensively against AI agents that route around their monetization layer, even when those agents could otherwise be complementary; expect friction as agents proliferate across e-commerce, travel, and other verticals.
Physical/operational infrastructure is becoming a strategic asset in the AI agent era — As software agents commoditize the "decision" layer, the differentiated value shifts to whoever controls fulfillment, logistics, or real-world execution — a signal for where durable moats may be built going forward.

6. Overlooked Insights

Frontier AI labs may have strategic incentive to "pace" development under the guise of safety — The brief mention of "Frontier Overhangs" suggests a subtler dynamic where slowing down publicly may serve competitive/strategic purposes (buying time to manage risks from prior model advances) rather than being purely altruistic — a nuance easy to miss amid typical AI safety framing.
Middleware players (like Expedia) are being squeezed from both sides — As AI agents and aggregators battle for control, intermediary platforms are quietly fighting to retain relevance, a dynamic that could preview broader disintermediation risk across other middleware-dependent industries.