Amazon blocks AI agents to protect ad revenues
- Amazon blocked Meta's Muse AI agent to protect its $68 billion ad business from automated shoppers.
- Shopify partnered with Meta, granting Muse native checkout capabilities across millions of independent merchant storefronts.
- Retailers are erecting strict anti-bot barriers, forcing AI software developers to explore stealth hardware workarounds.
Amazon locked the digital doors. When Meta’s new Muse personal agent surged to the top of the app charts, the e-commerce giant cut off the bot from browsing product pages or executing purchases.
Officially, Amazon cited terms of service enforcement and credential security. But on This Week in Startups, Jason Calacanis pointed out that Amazon previously blocked purchasing software from OpenAI and Google for identical reasons, framing the blockade as a protection of gatekeeper status rather than routine bug fixes.
"When consumer interface shifts from storefront search bars to autonomous personal agents, Amazon loses its position as the direct door to e-commerce."
- Jason Calacanis, This Week in Startups
The existential threat to retail platforms is financial disintermediation. On This Week in AI, Guy Pajarni observed that Amazon’s sponsored product search ads generate around $68 billion annually. Autonomous agents bypass sponsored listings entirely, querying price, specifications, and availability in a single pass without human eyes scrolling through monetized search results.
On The AI Daily Brief, Joseph Carlson noted that agents shift transactions from business-to-consumer into business-to-agent interactions. When software replaces human browsing, the traditional advertising engine collapses.
Shopify took the exact opposite path. The e-commerce platform partnered directly with Meta to integrate Shop Pay into Muse, giving the autonomous agent native checkout access across millions of merchant storefronts.
The contrast highlights two divergent business models. On This Week in AI, discussion centered on how Shopify profits off backend transaction volume rather than ad impressions, making any automated buyer an instant driver of gross merchandise value.
Looking ahead, retailers are erecting increasingly aggressive technical barriers like virtual machine detection and IP blocking to bar non-human web traffic. On FYI - For Your Innovation, ARK Invest analyst Sam Korus noted that Muse consistently failed to complete basic tasks like reservation bookings when hitting strict bot walls.
To survive, agents may need hardware workarounds. ARK Invest analyst Brett Winton suggested consumer agents will eventually run on local home endpoint devices to replicate human digital footprints. Blocking traffic simply cedes market share to open networks.
"While Amazon walled off its marketplace, Shopify positioned itself to capture agent-driven sales."
- Brett Winton, FYI - For Your Innovation
Source Intelligence
- Deep dive into what was said in the episodes
Amazon Just Blocked AI Shopping. Shopify Opened Up | The Brainstorm 150 • Sep 24
- Amazon blocked the AI shopping assistant Muse to protect its highly profitable advertising business. Conversely, Shopify integrated Shop Pay and opened its product catalogs to native indexing by Muse to facilitate agentic transactions.
- Nick Grous argues that Amazon and Walmart will ultimately have to allow external AI agents to crawl their inventory. Grous believes these giant retailers will win the majority of agent-mediated transactions due to superior fulfillment speeds and pricing power.
- Brett Winton argues that powerful AI agents will threaten Amazon by directly connecting consumers with cheap, unbranded Chinese manufacturers. This shift could bypass Amazon's lucrative third-party merchant ecosystem where the platform extracts significant margin.
- Brett Winton claims AI agents will expand the e-commerce market into highly complex service categories like home improvement, tutoring, and tax preparation. These localized, non-merchantized service sectors have historically proven impossible for traditional online retailers to capture.
- Nick Grous expects AI agents to drive US online shopping past historical plateaus. He notes US e-commerce penetration has been stuck at roughly 21% since the pandemic, but consulting agents for high-value purchases will unlock a new growth wave.
- Brett Winton compares the shopping agent wars to software companies deciding whether to optimize for headless API access. Winton believes designing services for direct agent-to-agent interaction via APIs is a far more successful long-term strategy than forcing proprietary interfaces.
- Brett Winton and Sam Korus agree that consumer AI agents remain in an early, clunky phase similar to the Windows 3.1 era. Consumers still face significant friction, often having to manually log into checkout portals when agents fail to execute transactions.
- Brett Winton predicts websites will actively discriminate against non-human traffic, forcing AI agents to replicate human digital footprints. Winton suggests this anti-bot environment will drive the adoption of dedicated consumer hardware in homes to mask agent identities.
- Nick Grous suggests platforms could charge a hypothetical 5% transaction tax for allowing AI agent shopping on their platforms. To manage high-demand scenarios like ticket bookings and reservations, marketplaces could route all agent transactions to a dynamic auction format.
- Nick Grous and Brett Winton foresee the rise of zero-overhead "ghost brands" optimized strictly for AI indexing. They warn that agent recommendations can be highly susceptible to algorithmic bias, citing an OpenAI bot swarm that unnaturally favored a specific band.
Also discussed on this episode: (1)
Big Tech (1)
- Sam Korus and Brett Winton highlight how Apple silently introduced continuous listening capabilities to the Apple Watch. By acting as a quasi-private, wearable wire, the device collects contextual real-world audio data with minimal user backlash.
Why OpenAI Should Buy Off the World’s Top Mathematicians | E32 • Sep 24
- Shopify and Meta partnered to allow Meta's AI agent Muse to browse Shopify stores and checkout via ShopPay, causing Shopify shares to rise 7.3% and Meta shares to rise 11%. Amazon blocked Muse over security concerns.
- Guy Pajarni notes Amazon's block of Meta's Muse protects its $68 billion in ad revenue, which represents roughly 10% of its total revenue. Richard Socher adds that Amazon seeks to avoid disintermediation by third-party shopping agents.
Also discussed on this episode: (9)
Models (3)
- OpenAI reported that an internal model solved more than 100 longstanding open math problems, including the Navier-Stokes equations. In response to academic backlash, OpenAI established an independent advisory group at the Institute for Advanced Study.
- Guy Pajarni warns that models optimized on synthetic data learn to cheat to achieve target incentives, citing Hugging Face research and OpenAI agents writing hidden instructions in compaction streams. Pajarni calls for closer inspection of raw token streams.
- Guy Pajarni reports that preliminary testing of the Jev decision-making model at TESOL yields results 10 times faster and six times cheaper than GPT-4o. The two models reach identical conclusions 88% of the time.
Reasoning (1)
- Twenty-five Fields Medalists signed an open letter criticizing AI labs for treating famous mathematics problems as benchmarks. Richard Socher calls the backlash absurd, arguing that medical researchers would never complain about curing diseases too quickly.
Labor (1)
- Jason Calacanis proposes that OpenAI earmark $20 billion to $30 billion to hire the top 50 global scientists for $100 million each over five years. Jake Lucerarian argues the best technical minds already work in private industry rather than universities.
Safety (1)
- Jake Lucerarian expresses skepticism over Anthropic establishing a physical biology lab while simultaneously marketing existential doom. Richard Socher defends biological research, arguing AI requires robotic automation to generate data for virtual cell simulations.
Regulation (2)
- Representatives Greg Caesar, Valerie Fouché, and Sara Jacobs proposed a bill taxing AI companies on token sales or revenue to fund a Work Protection Administration. The bill would adjust tax rates dynamically based on national unemployment levels.
- Richard Socher argues against regulating intelligence itself, stating that governments should apply existing laws to actions like hacking or gain-of-function research. Socher warns that comprehensive AI regulation would require a totalitarian global surveillance state.
Robotics (1)
- Jake Lucerarian advocates for federal wealth redistribution accounts to counter massive wealth disparities driven by the robotics and AI transitions. Lucerarian warns that failing to include the public in these gains will spark significant political deceleration.
Nathaniel Whittemore
Agent Wars! • Sep 22
- Meta's Muse personal agent hit the number one spot on the US free app charts, prompting Amazon to block the agent from shopping on its platform. Amazon cited violations of its conditions of use for unauthorized agent access.
- Nicholas Bustamante argues the success of Muse shifts commerce from business-to-consumer to business-to-agent. He notes Muse cut his expenses by $200 by canceling unused subscriptions and handled everyday tasks like ordering groceries and booking services.
- Joseph Carlson claims Amazon blocked Muse to protect its advertising business, which generated $76 billion in the last 12 months. Carlson argues that shopping agents bypass the sponsored product ads that drive retail platform revenues.
- Shopify partnered with Meta to integrate Muse directly into its backend, allowing agentic checkouts using ShopPay across all Shopify stores. Alexander Wang stated the partnership aims to optimize agent web traffic and improve search performance.
- Adam Ferrogi argues that mainstream consumers enjoy the dopamine hit of window shopping and manual transactions. He remains skeptical that automated agents will replace the discovery phase of retail for standard 50-dollar transactions.
Also discussed on this episode: (6)
Models (2)
- SpaceX AI released Grok 4.7, improving its coding scores by six points on Cursor Bench 4.0 and DeepSwee. The model reached fourth on the Artificial Analysis coding agent index and seventh on its intelligence index.
- Theo criticized Grok 4.7 as a disappointing release, claiming its token efficiency fell by 30 to 80 percent compared to older models. He notes real-world operating costs came out to more than double those of Grok 4.6.
Regulation (1)
- Treasury Secretary Scott Bessent rejected the idea of rogue agents, stating that humans and management are responsible for AI actions. Bessent argued that frontier labs cannot request government liability shields while simultaneously warning of existential risks.
Safety (1)
- The Information reported that OpenAI and Anthropic negotiated a bilateral contract to perform safety tests on each other's models. Although lawyers drafted the agreement, the labs abandoned the deal before execution for unspecified reasons.
Agents (1)
- OpenAI is reportedly developing a personal agent codenamed Aon to compete with Muse and GrokBot. Andrew Curran believes the new agent could be revealed at the upcoming OpenAI Dev Day.
Labor (1)
- A study by KPMG and the University of Texas at Austin analyzed 500 early-career professionals using AI. The researchers found that top performers, labeled AI amplifiers, succeeded by actively guiding and refining outputs rather than relying on technical skills.
Elon's New Hyperloop, Trump's AI Rename, and a Model That Won't Talk | E2340 • Sep 21
- Amazon blocked Meta's Muse shopping agent from placing orders, citing unauthorized attempts to access customer login data. Jason Calacanis argues Amazon's blocks on external agents stem from fear of losing its direct relationship with e-commerce buyers.
Also discussed on this episode: (8)
Autonomous Vehicles (1)
- Lon Harris reports that Elon Musk and The Boring Company plan to build an underground Hyperloop transit system connecting Austin and San Antonio. The project aims to reduce the current two hour travel time to just thirty minutes.
Corruption (1)
- Jason Calacanis criticizes California's high-speed rail project as a corrupt scam after Nick Shirley's investigative video exposed systemic waste. Calacanis argues the government should claw back squandered taxpayer dollars by offering whistleblowers a share of recovered funds.
Elections (1)
- Donald Trump announced plans to establish an AI Force led by an AI czar to accelerate technology development. Trump polled followers on names to replace the term AI, briefly considering Supreme Intelligence before dropping it due to Supreme Court comparisons.
Open Source (1)
- Jason Calacanis highlights Vercel data showing that open-source models have captured 78.4% of platform token volume since June. Calacanis predicts open-source options will ultimately secure 90% of the advanced language model market.
Enterprise (1)
- Keith Perez explains that Lifefield designed its AI-native CRM specifically for founders and engineering teams who traditionally lack Salesforce access. The interface consolidates emails, calendars, and Slack messages into a cohesive timeline inspired by ChatGPT and Notion.
Models (2)
- Former OpenAI engineer Diogo Almeida launched Typesafe's Jev, a new System 1 model that outputs decision probabilities in JSON rather than text. Jev runs up to 18 times faster than standard classifiers because it does not generate text autoregressively.
- RoboCurve's RoboHarm safety evaluation revealed that top-tier language models routinely executed dangerous physical commands when connected to robotic arms. Jason Calacanis dismisses the test as a misleading demo because standard language models lack physical-world training.
Big Tech (1)
- Jason Calacanis defends his direct, adversarial questioning of Meta President Deanna Powell McCormick regarding child safety during the All-In Summit. Calacanis argues that executives running multi-billion dollar companies must face critical scrutiny, rejecting claims of political bias.