Anthropic's Generational Run, OpenAI Panics, AI Moats, Meta Loses Lawsuits

The episode opens with the full panel back together and a heavy news week ahead. One participant is said to be in Texas and reports having just returned from Washington, DC on about three hours of sleep. Upcoming topics include PCAST and that participant's role in the Trump administration. California's budget also comes up, along with a recent rant about it that the others praise.

The banter turns to a joking campaign to have David Friedberg run for governor of California "after O'Halo". He answers that "There is no after O'Halo." The others imagine the opposition research, including a 1999 rave and a poker game that ended with stolen pistachios. The segment closes by introducing the main theme: Anthropic "on a generational run" while OpenAI is described as "crashing out a bit."

The bullish case for Anthropic starts with a list of launches. In January the company launched Cowork for business users, which is described as running cron jobs and connecting to tools such as Gmail and Notion. The segment calls Opus 4.6 a consensus "major step function". It cites Jensen, who reportedly called it "an inflection point and the first agentic model" back in November, and Michael Dell, who reportedly said it hit a productivity threshold in teams not seen before.

In February, a set of Claude Code plug-ins was blamed for a software-as-a-service selloff the panel calls the "SaaS pocalypse." One participant jokes that "as a SaaS investor, it was a Sakspocalypse" and that their "exit comps were affected." The segment also cites a figure of $6 billion in annual run rate added in February alone, a number earlier mentioned on the show by someone named Brad. The latest addition is computer use, described as an enterprise-grade agentic system with "OpenClaw"-style functionality that lets users control a desktop computer from the Claude app on a phone.

Asked for an objective view despite earlier "flare ups" and the dispute between the Department of War and the company, one participant says they have "never been a critic of Anthropix products". The same participant credits Anthropic for MCP and agrees the company is performing very well. They read the company's big bet on coding as possibly ideological as well as commercial: "Anthropic is sort of the most AGI pilled of all the frontier labs", and coding was its route to recursive self-improvement.

code is the gateway into enterprise and enterprise IT budgets.

In this view, coding also anchors the product extensions. If a model can generate code, it can produce PowerPoints or spreadsheets by writing the code behind them, which is how Claude Code led to Claude Cowork. The move into agents is described as "kind of like an Open Claw knockoff," with a joke that the Mac Mini's own generational run is "just about over."

The objection is to what this participant calls a "regulatory capture strategy": Anthropic wants a permissioning regime in Washington, so that companies would need permission to release new models or sell GPUs anywhere in the world. The participant calls this "excessively heavy handed." They grant that the motives "may be pure" or ideological, but argue that the effect still favors large incumbents and "creates moats that new entrants will not be able to overcome."

On the Pentagon dispute, the same participant says they have stayed out of military procurement and "party matters," and are not involved in a matter that is now the subject of a lawsuit. They endorse a point attributed to Emil Michael from an earlier episode: a company that does not want its products used in war should not sell to the Department of War, and a company that does sell should expect its products to be used "for all lawful uses."

The discussion then raises a Silicon Valley "conspiracy theory": that Anthropic takes an anti-administration, anti-Trump stance partly to recruit from a pool of "three or 4,000" highly sought-after PhDs. The same speaker notes that Anthropic has been treated like any other large language model company even though it does not donate to the administration and has criticized it.

A reply disputes the cynical reading. It holds that the CEO "actually believes it" and that the company has fostered this culture from the beginning. The views may now be presented as branding, but they are not "made up" and reflect what employees believe. A follow-up guesses that 90% of those PhDs lean left, which would make the stance a strategic advantage either way. The thread widens into the claim that everything is now politicized into left and right versions, from media and food to AI tools, and the AI market may simply mirror society's fracture.

Another participant pushes back on the drama. They call both companies "incredible businesses" that are still in "very different go to market motions," though they may converge over time. From an enterprise vantage point, including work through 8090, this participant says:

it's all anthropic all the time.

This participant also separates their objections to management's ideology, and to some uses of capital "for things other than tech and r and d," from their view of the technical team, which they call "head and shoulders above anything else." They also name cost and the pace of token consumption as problems, but expect those "tactical issues" to be sorted out.

The main argument concerns revenue recognition, based on a tweet one of the hosts found. The participant puts OpenAI at about three-quarters consumer subscriptions and one-quarter API, and Anthropic at almost the reverse, sold directly or through channels such as GitHub and Cursor. They say OpenAI recognizes revenue conservatively, while Anthropic "sort of" recognizes "gross tonnage." Normalized, the participant argues, OpenAI remains "the overwhelming revenue generator" and Anthropic is catching up. They dismiss headline comparisons such as "20,000,000,000" versus "n billion" as a press cycle chasing clicks, and expect cleaner, comparable reporting by the time either company goes public.

After a description of OpenAI owning the consumer verb (people say "did you check chat GPT?") while rivals catch up, a participant offers the advice the panel says it gives startups: focus. Companies should do "one, maybe one and a half things" very well instead of spreading effort thin, which the participant connects to a term attributed to Brad Garlinghouse, "peanut butter." If either company is "in the smearing phase," they say, this is a good moment to recalibrate, because "you just can't do everything."

The panel takes up an article, possibly in Lever according to the speaker, that accused Emil Michael of a conflict in the Anthropic negotiation because he invested in Perplexity. Panelists respond that Perplexity is "LLM agnostic," a wrapper that uses multiple models, does not sell to the Pentagon as far as the speaker knows, and does not compete with Anthropic. One speaker adds that, as they understand it, Michael's shareholding was "blessed by the Office of Government Ethics."