Saturday Night Live parodied Dario Amadei, a private AI-lab leader, in a sketch that mocked the difficulty of discussing AI’s potential dangers, calls for legislative safeguards and possible benefits. The sketch’s character says there is about a 10% chance that cancer will no longer be a problem for anyone in 10 years, then describes AI as “a tool for building weapons.”

The parody reflects how AI leaders and the debate around the technology have entered wider public view. Amadei’s appearances on major news networks were cited as part of the reason he had become recognizable enough to serve as an SNL character.

An analysis of AI leaders’ public communication argued that a sober, restrained tone can signal that serious risks are being taken seriously. It criticized Dario for smiling while discussing devastating outcomes, while also describing his thinking as exceptionally clear.

The analysis contrasted that style with Mark Zuckerberg’s more upbeat approach, which may appeal to some people but was judged unlikely to resonate with researchers pushing the AI frontier.

Andrew Curran reported that President Trump personally invited Dario Amadei to a private dinner on Sunday. It was described as their first one-on-one meeting. Amadei had appeared to be absent from an earlier White House dinner with Xi Jinping and top technology leaders; Trump held the private dinner to hear his side.

A separate dinner in Washington for AI leaders and the White House was planned for Tuesday night. The source described AI meetings as taking place nearly every week.

A proposal published through the University of Cambridge and signed by academic and industry researchers calls for comparable measures of how much AI contributes to labs’ research and development—for example, the share of code or spending that is AI-enabled.

The proposal says differing measures make it difficult to compare labs’ progress.

A Wall Street Journal article reports that some early Anthropic employees are considering buying remote land in the United States to relocate to if AI goes seriously wrong. They also practiced doomsday scenarios and preparation plans in a private Slack. The wider AI safety community has discussed iodine pills, remote islands and electromagnetically shielded desert bunkers.

An isolated, well-stocked bunker may not be secure if its owners lack ties to the local community; in a crisis, it could become a “loot drop.” A separate policy concern was a stated preference for regulation rather than facing a self-defense situation involving rogue agents. Rogue Waymo vehicles were raised only as a hypothetical example.

Jan Tollen, identified as Skype’s founder and an early investor in AI labs, is described in a journal piece as a primary funder of much of the AI-doom and effective-altruism scene.

He warns against racing to build technology that could end humanity. The piece also recounts a debate about machine consciousness at 40,000 feet over flutes of champagne.

Instinct was reported to have reached $1 billion in transaction volume in about six months, but it is unclear whether the figure is annualized or cumulative to date. The invite-only service reportedly launched to about 200 friends and family, grew 10% a day with no marketing spend, and had invitations listed on eBay for around $300. About 40% of users reportedly shared a personal credit card within three weeks. A projection—not a current user count—said 5%–8% daily growth while new compute came online could take Instinct to about 100 million users; compute demand was described as roughly doubling weekly.

Instinct’s own A/B tests reportedly matched Opus 5 performance at a fraction of the cost. Last-minute compute purchases were said to cost three to four times more, while background work could run on setups described as three, five, and eight times more efficient on the same compute. The service was also described as retaining about 80% of users who connect at least one piece of sensitive information. It has reportedly called founder Noah Shin three times, including to alert him that a task needed signing within five minutes. The adoption assessment is that most people outside tech have yet to experience AI completing useful tasks, which can feel magical.

Instinct plans to take a cut of purchases, but transactions currently run through third-party cards and do not yet generate revenue for the company. A proprietary card and booking fees were discussed as possible revenue sources. Travel was cited as 50% of transaction volume—not prompts—though 15% was also mentioned before the 50% figure was reiterated, leaving the exact share unclear.

Noah Shin, founder of Instinct, thinks agents could wipe out the 70% of a hypothetical app’s revenue that comes from ads, while making its 30% from product sales much bigger because people may buy more when purchasing takes no effort. This is a possibility, not a measured outcome; whether higher sales would offset lost ad revenue is unclear.

The potential benefit may depend on the shopping task: buying through a Shopify store can already take little effort when shoppers know what they want, while agent help could be more useful for tedious purchases where brand matters less, such as finding a well-reviewed set of cable-management accessories.

Personal AI agents could negotiate for restaurant tables and match diners with venues based on preferences and special occasions, rather than relying only on who books first. But restaurants may continue to reward advance planning and personal relationships; some high-demand tables are held outside booking systems. One forecast is that restaurant-side systems will adapt to agent inquiries, while agents could help fill seats more efficiently, benefit smaller restaurants and bring more people to more venues.

The shift could also produce repeated calls when online listings show no availability, creating what one prediction described as “fake demand” and prompting more filtering. A separate, explicitly uncertain possibility is SeatGeek-like pricing: top restaurants might charge as much as $2,000 for a reservation, while others might price food 1% above cost to fill seats.

Tate Hacker predicts that everyone will be using a full-time “assistant of compute” like Instinct’s personal agent within a year. Nearby context describes personal agents for tasks such as commerce and flight booking, and says the agent already has essentially a computer, phone number and email.

The remark also refers to a parking ticket and a car Instinct took care of, but does not specify exactly what Instinct did.

Meta announced Meta Enterprise Platform, a new business pillar intended to help companies use AI. Its initial offering will bring businesses and developers Meta’s models, agents, infrastructure and tools, including Muse Agent, Meta Business Agent, Muse API and Muse Code. Shantanu CJ Desai will join as Chief Enterprise Platform Officer, reporting directly to Mark Zuckerberg.

One speculative scenario is that the unit could quickly grow from $0 to $10 billion in annual revenue, without initially specifying how much comes from tokens, inference, compute or services. Selling compute could provide another outlet for Meta’s infrastructure, but may compete with the resources needed for its own models and products.

Adidas’s store uses anti-bot software, which reportedly got in the way when Codex was used to research the store—even though the user was trying to give the company business.

The incident highlights a challenge for online retailers: maintaining protection against harmful bots while supporting legitimate consumer agents.

The chief economist at Apollo warned that mass adoption of AI agents optimizing users’ investments could trigger a bank run. Gary Gensler, during his final years as SEC chair, also warned that algorithmic decisions converging at scale could undermine market stability; similar choices by personal finance agents could potentially cause a massive flash crash. Their similar actions could also affect areas beyond markets by removing economic friction, one analysis argued.

Adoption may be gradual: one estimate put current use of personal AI agents in the single-digit millions, and wider use would require people to authenticate, link bank accounts and build trust over time. Some banking services already proactively move customers’ funds into money-market accounts or T-bills. If agents spread over a couple of years, banks and money-market funds may adapt, potentially limiting the disruption.

MongoDB shares were down 16% after falling 26% earlier, then recovering somewhat.

The remarks did not establish what caused the move.

A pitch to investors for Scale AI-related SPVs presents the company as a provider of software used to train major AI models, claiming that about 90% of the world’s leading generative AI models use it. The excerpt gives no terms for the SPVs or outcome of the offer. It also says Meta put $14 billion into Scale AI a few months earlier, despite Meta’s substantial spending needs for data centers and its own AI system.

Scale’s remaining business is described as still generating revenue and doing deals, while substantial talent went to Meta and the founder was leaving. The pitch discussion says Scale lost many customers after the deal, with some concerned about an exclusive relationship with Meta. Early investors received a dividend but still own shares. Describing the deal as regulatory arbitrage, and suggesting remaining shareholders may see little future for Scale, are opinions rather than established outcomes.

Sonnet 5.5, the second model in the Claude 5.5 family, is out.

It is described as a clear upgrade over Sonnet 5, running 30% faster and costing up to 30% less for most work. It also shows good results in agentic coding.

Plexo Capital’s strategy grew out of work at GV, where its founder focused on marketplaces and mobile and the firm invested in about five emerging micro-VC funds, typically backing pre-seed or seed companies.

GV also helped those managers with fundraising.

GV operated at the financially focused end of the corporate-venture spectrum, as a separate unit under Alphabet’s Other Bets umbrella.

Its investors sought financial returns like a traditional venture firm; Alphabet introductions could be made when useful, but the priority was identifying strong companies.

Product-management roles can offer strong preparation for founding a company and evaluating pre-seed and seed investments, by bringing together work on engineering, feature priorities, marketing, distribution, business models and monetization.

The investment process uses a similar playbook: assess the problem, existing solutions and their shortcomings, and how a product can reach users.

Plexo Capital’s founding and managing partner says AI can help the firm handle non-investing work without building a large operations team. At GV, the partner had finance, legal, operations and marketing teams; running an independent firm put those responsibilities on their plate, while limited management-fee income could make a large team difficult to afford.

The partner recalled telling LPs in fall 2023, while looking at Anthropic, that AI doing the work of one junior analyst would be valuable. They now believe AI can do the work of a whole team. Deep-research tools can also help users investigate a technology and its market and get up to speed “10 times faster,” one assessment said, although the resulting writing may still contain “AI slop.”

Some funds that raised in 2024 aimed to own 5%–10% of companies at pre-seed or seed, but almost all of those managers are now happy to get 3% in a round, according to an example in the source. Sharp markups and fast follow-on financing can force managers to revise ownership targets even within a single investment period.

Smaller funds also face a growing challenge in maintaining their pro rata share as companies move quickly through financing rounds. In the past, managers used SPVs and later dedicated follow-on funds to help preserve those stakes. Both approaches are now more difficult as rounds grow rapidly and larger firms enter at earlier stages.

An investor argued that the best GPs just starting their careers are more technical than their counterparts were a decade ago—a change they see as especially important in the age of AI.

Experience tackling problems inside technology companies can help investors recognize technical limits in existing solutions and spot what is missing, such as more computing power or an unresolved architectural problem. Time at major technology companies can also help them assess employee and team quality and understand how strong organizations scale.

A venture firm’s first SPV was used to invest in Anthropic’s Series D, which closed in January 2024 and was led by Menlo.

A GP who had previously worked with the firm on Reddit brought the deal. The firm was seeking more AI exposure, but the $18 billion valuation was far higher than any it had previously considered.

Venture GPs are clarifying whether they focus on building companies in inception rounds or act as financing partners, and which company stages and needs they serve. One investor said this level of precision is necessary today, unlike five or ten years ago.

A fund’s edge may come from technical expertise that helps identify opportunities, a focus on companies at a particular point in product-market fit, or specialized networks. Examples included access to a network of product managers and a GP partner with a background in PR and company positioning.

Startups can now reach major scale in about 10 months, compared with trajectories that once allowed founders roughly 10 years to develop the skills needed for high-profile CEO roles, including handling demanding boards and public appearances.

Higgsfield was cited as the fastest company to reach $1 billion in revenue, though the claim was qualified as something the speaker had just seen.

Consistency in a GP’s choice of sectors, problems and founder profiles can help LPs assess a manager’s investment approach, while distinctive GP networks may bring less-overlapping deal flow.

When considering vice-related investments, LPs also need to account for restrictions in fund LPA terms and the values of individual LPs.

Zack London, known online as Gossip Goblin, says his feature-length film is set to premiere December 4. His team has spent the past 8–9 months making it, after building a sci-fi universe through short-form social-media videos that attracted enough interest for him to leave his day job, open a small studio and hire a team.

London says the film is not made by simply prompting an AI tool: he and a co-writer wrote the script, and the production used a dozen voice actors, musicians, Foley artists and a post-production crew. He has been making AI videos for about four years and says the tools reached an inflection point roughly a year and a half ago, making watchable work possible. He describes the process as arduous, though faster and leaner than producing a sci-fi epic of similar scope in the past.

Zack London said he thinks there is a 90% chance humanity will survive.

Asked whether he is a doomer or a techno-optimist, he said he can be both, brought up transhumanism, and stressed that he is not a doomer. He did not specify a time horizon or the threats behind his estimate.

A user was blocked while trying to buy something on Adidas.com with Muse.

The reason was not confirmed; the site’s defenses against more harmful bots, such as sneaker bots, were suggested as a possible cause. The example highlights how anti-bot systems may also obstruct legitimate personal agents.

Personal AI agents could make it easier for consumers to request refunds, dispute payments and handle other bureaucratic or legal processes.

One analysis predicts this may create more claims in areas where merchants currently benefit from customer apathy or lack of information, and could spur demand for software to manage disputes. The scale of this effect has not yet emerged.

A current estimate puts the cost of serving a personal AI agent at about $1,000 per user per year, a level described as prohibitive. For a free service funded by 3%–5% transaction fees, costs would need to fall meaningfully: the average American was said to spend $5,000–$6,000 a year on online travel and commerce, while a 3%–5% fee was noted as difficult to make up the service cost without much higher spending or a high take rate.

The view expressed was that compute costs will decline, potentially quickly. Astra was described as capable at computer use but currently very expensive; computer use with Jed was described as 100 times cheaper and faster today, as a possible route to faster cost reductions.

A dedicated financial tool for personal AI agents is one possible starting point: a hypothetical credit card from Instinct could monetize agents’ purchases without requiring opt-ins from every retailer or service provider.

The idea was raised amid skepticism about pitches for agents to use stablecoins. However, the view expressed was that card transaction fees alone may not be enough.

Tesla has postponed its Roadster update event from October 7 to October 15.

The event can only be held outdoors, and the company rescheduled it after tracking forecasts of severe weather.

San Francisco Compute says it helps turn resources into cloud compute capacity and helps new providers sell and finance that capacity through a market intended to reduce their risk.

The company argues that a reliable compute market needs an independent operator with physical responsibility for infrastructure—from managing data centers, and sometimes building them, to standardizing how capacity is delivered. It says index prices and cash-settled futures alone are disconnected from the underlying compute.

Gstaad Guy says his in-character videos satirize extravagant lifestyles, drawing humor from how consumption signals status. He has a team, but says he feels celebrated by the term “creator”; he believes some creators prefer “media company” because they feel insulted by the creator label.

The satire draws on contrasting status markers: in startup-oriented circles, people may be judged by the companies they build, fundraising and team size, while in some European circles the focus may be a watch’s uniqueness, history and maker—not simply its price.

Gstaad Guy’s Constance and Colton represent contrasting impulses in luxury consumption: heritage and discretion versus hype and status. The creator says the characters use satire to tell the “what” and “how” of that world, and can reflect opposing desires within the same wealthy consumer.

He says he now rarely uses Colton because audiences are saturated with similar characters, while Constance remains comparatively novel. He links that appeal partly to the privacy of people who fit the archetype and their tendency not to post their lifestyles as much. For the “why” behind luxury, he breaks character: he felt it would be disrespectful to address the motives of people—such as hoteliers, restaurateurs and craftspeople—who put serious love and passion into their work while speaking as a character.

Gstaad Guy says he started the account about eight years ago after a private video spread on WhatsApp and someone stopped him on the street to say they had seen it. Before Reels launched, Instagram paid selected video-native creators, including him, to use Reels instead of IGTV. He says he has followed his audience rather than the algorithm and still aims to serve the account’s original niche of private, wealthy followers, despite now having millions of followers.

He says he rarely uses the Colton character now because audiences are tired of that archetype, while Constance remains novel partly because people who fit it are private and do not post their lifestyles as much. He tracks whether specific audience types engage with videos made for them, calling it a failure if the intended person does not engage—even if a video gets millions of views. He says brands work with him because they struggle to reach that niche audience; the strongest current business demand is in the US, where brands want him to bring a European touch and storytelling to their businesses. He attributes the page’s rapid growth to Americans’ curiosity about Europe.

AI may deepen the divide between efficiency-focused businesses and luxury hospitality, according to an argument that automation can make already optimized companies even more optimized while making exclusivity more pronounced. The same view holds that hotel guests will still value a human manager who can support their stay, and that craftsmanship and creative direction remain central to fields such as watchmaking and fashion.

A video was said to claim that thousands of small European hotels may need new owners within the next 10 years because there are no heirs, and proposed rolling them up. The concern is that scaling these family-owned businesses could destroy what made them distinctive: a model designed to stay in business, not maximize extraction and growth. Il San Pietro in Positano was cited as an example of enduring loyalty: more than 70% of its guests had stayed for over 10 years and more than 50% for over 20 years; new guests came by referral. The argument also holds that private equity usually ruins luxury without creating value or saving jobs, and predicts that AI will ruin a lot of luxury because, in this view, luxury is meant to remain unoptimized.