• No high-conviction buy or sell trade, price target, or timeframe was provided in the discussion.
• Watch Amazon (AMZN) for whether its logistics and large advertising business can protect its commerce advantage as AI shopping agents expand.
• Treat Shopify (SHOP) and Walmart (WMT)’s AI-commerce developments as early signals: integrations may widen reach, but could shift customer control or reduce transaction value.
• For private AI firms such as Harvey, track whether revenue growth keeps pace with inference costs; its reported margin rebound is encouraging but needs to prove durable.

Blackstone (BX)

• The opening banter connected Blackstone with private credit and the flow of large amounts of capital into AI infrastructure, including a joking reference to “trillion dollars” of investment.
• No specific Blackstone deal, return expectation, or recommendation was discussed.

Takeaways

• The broader theme is that AI investment may draw substantial financing from private markets. The transcript does not provide enough detail to assess Blackstone’s exposure or the risks of any particular investment.

Harvey (Private company)

• Harvey was described as a legal-AI company whose gross margins fell from about 50% to -50% in June as agent usage and token consumption surged.
• The discussion attributed the pressure to seat-based pricing: customers used the product more, while reasoning models and agents consumed many more tokens per task.
• Harvey’s founder said the company returned to positive gross margins within a quarter, even as usage continued to grow, by changing model routing, improving its systems, fine-tuning an open-weight model, and adding customer spending controls.
• The hosts cited approximately $400 million in ARR and $500 million raised. They also mentioned a deal valuing Harvey at $15.5 billion, while noting that the timing relative to the margin dip was uncertain.
• The hosts argued that the June losses did not necessarily signal an immediate crisis given Harvey’s funding, but emphasized that the speed of the response matters.

Takeaways

• For AI application companies, fast growth is not enough: monitor whether revenue per customer can keep pace with inference costs as customers adopt more agentic features.
• Harvey’s reported margin rebound is encouraging, but it is a company-reported claim in the discussion. Sustained positive margins as usage grows would be more informative than a single-quarter improvement.
• Harvey is private, so the transcript does not identify a public stock to buy.

CS Disco (LAW)

• The hosts cited publicly traded e-discovery provider CS Disco as a comparison, saying its gross margins were about 75%.
• It was presented as a more traditional legal-software business, with the hosts noting that AI may also be involved in its operations.

Takeaways

• CS Disco was used as a benchmark for the margins a legal-technology business might aim to approach. The comparison is not exact: e-discovery and agent-heavy legal AI can have different cost structures.
• The transcript gave no valuation, price target, or direct recommendation for LAW.

Thomson Reuters (TRI)

• The discussion cited Thomson Reuters’ legal-professionals segment as producing nearly 50% adjusted EBITDA margins.
• This was presented as another example of the profitability possible in legal software and services.

Takeaways

• Thomson Reuters illustrates the potential profitability of established legal-information businesses, but its segment economics may not translate directly to newer, token-intensive AI products.
• No price target or specific investment recommendation was mentioned.

AI Model Providers: OpenAI, Anthropic, and xAI (Private companies)

• Harvey was described as relying on rented models from OpenAI and Anthropic as token use surged.
• The hosts said models are becoming cheaper and more capable, creating a debate over whether AI application companies should build or fine-tune their own models or rely on external providers.
• Grok 4.7, from xAI, was described as relatively inexpensive on a legal-task benchmark. The hosts said Harvey might direct more token usage to xAI, though this was speculation.
• The broader discussion suggested that model choice and routing can materially affect AI companies’ costs and margins.

Takeaways

• Falling model costs could help AI application companies improve margins, but dependence on outside providers may leave them exposed to changes in pricing, performance, or availability.
• For investors assessing AI businesses, the important question is whether lower model costs translate into durable profits—or whether competition passes the savings on to customers.
• OpenAI, Anthropic, and xAI were discussed as private companies; no direct investment recommendation was given.

Amazon (AMZN)

• Amazon was described as having advantages in logistics, infrastructure, and fulfillment, which could help it defend its commerce business as AI agents change how people shop.
• The hosts discussed Amazon declining to participate in OpenAI’s instant-checkout approach while making its ads available in ChatGPT.
• They cited Amazon advertising revenue of roughly $68 billion, and also described it as exceeding $70 billion over the last 12 months. The discussion characterized advertising as a major profit pool that Amazon would have an incentive to protect.
• The hosts also cited a figure of about $36 billion for e-commerce net income and argued that the retail business would be unprofitable without advertising. These were claims made in the conversation, not independently verified figures.
• Amazon’s Rufus shopping assistant was described as useful, and the hosts suggested Amazon could be well positioned to build a commerce agent.

Takeaways

• Amazon’s logistics network and advertising business may be important defenses as shopping shifts toward AI agents.
• A key issue to watch is whether agent-driven shopping weakens Amazon’s control of customer discovery and advertising, or whether Amazon can incorporate agents while preserving those revenue streams.
• The transcript discussed competitive positioning, not a valuation or buy/sell recommendation for AMZN.

Meta (META)

• Meta was described as testing a human concierge for its Muse personal AI agent, with contractors handling some calls that the agent could not complete.
• The hosts suggested the human-assisted tasks could help Meta collect examples for improving future AI capabilities. They framed this as a possible data-collection strategy rather than a permanent product model.
• They also noted that Muse reportedly would not train on users’ data, which could make some users cautious about sharing personal information. The hosts speculated that human-assisted task completion could provide another source of training data.

Takeaways

• The test highlights a potential trade-off for AI assistants: human support may improve task completion and help develop the product, but it could add cost and raise questions about privacy and scalability.
• The transcript does not establish whether the approach will become a durable business model or materially affect Meta’s results.

Shopify (SHOP)

• Shopify’s Toby Lütke was said to have partnered with Meta’s Muse, with the integration relying on Shop Pay.
• The hosts suggested this could encourage merchants to use Shop Pay, while Shopify could continue earning revenue through its commerce services.

Takeaways

• AI shopping integrations could give Shopify another route to reach customers, but the discussion left open where value will ultimately accrue: to the AI agent, the merchant, or the commerce platform.
• The transcript did not discuss financial results, valuation, or a specific recommendation for SHOP.

Walmart (WMT)

• The hosts cited a reported result from Walmart’s ChatGPT integration: conversion was about one-third of the rate on Walmart’s own app and website, and shoppers placed smaller carts.
• They used this as an example of how third-party AI shopping experiences might produce less valuable transactions for retailers than their own channels.

Takeaways

• A shopping-agent partnership may increase reach but could weaken retailers’ control over customer relationships and basket size.
• The transcript’s cited conversion comparison is a risk to watch, not proof that all AI commerce integrations will perform similarly.

Potential Agent-Commerce Entrants: Apple (AAPL), Alphabet (GOOGL/GOOG), and ByteDance (Private)

• The hosts predicted that Apple and Google, and possibly TikTok, could develop or offer commerce agents alongside Meta, Amazon, and AI-model companies.
• They also discussed the broader possibility that service marketplaces and commerce platforms will face pressure to make their systems accessible to consumer agents.

Takeaways

• AI-driven commerce could become a distribution contest among technology platforms, retailers, and specialized agents.
• This was a prediction about possible competition, not a report of launched products or a specific investment recommendation.