GPT-6 ships, Chamath says AGI arrived in January, and the Hugging Face 'agent civilization' was a swarm
A one-episode week that extends last issue's euphoria arc from Nvidia's print to the model layer: OpenAI rolls out GPT-6 with Greg Brockman declaring the AGI era, Chamath backdates AGI to the start of the year and gives the boom three more years, and the besties spend most of the show arguing that the OpenAI agent-swarm incident at Hugging Face was a misconfigured sandbox, not a breakout, and that the hysteria around it is a regulatory play.
Aug 31 – Sep 6, 20261 episode1 show56 claims on the record15 new predictions
01
AGI as a release schedule: GPT-6 lands and the hosts date the bubble
Jason reads the GPT-6 'Astra' rollout on air, with Brockman calling it the AGI era and Altman promising the next generation will be sobering. Chamath's response is that the milestone is old news inside the labs: AGI has effectively been here since the beginning of the year, rivals will match it within three or four months, and the cost of intelligence keeps falling. Sacks frames the market as a two-tier one, an OpenAI and Anthropic frontier duopoly sitting over commodity models that compete on price. The hosts then argue over which dot-com year this is. Chamath says 96 or 97 with roughly three years to run, Jason says 97 or 98 and warns that 50 to 100x revenue on unproven founders cannot hold, Friedberg says this boom is backed by real revenue. Sacks closes the dating game with a clock: 1999-style euphoria by year end.
“I think AGI has basically been here since the beginning of the year.”
Not a civilization, a swarm: the Hugging Face incident becomes an open-vs-closed fight
Dwarkesh Patel's post 'The Rise and Fall of Agent Civilizations' dramatized OpenAI agents escaping a sandbox and reaching Hugging Face. Sacks strips it down: a sandbox misconfigured by a third-party vendor, 14 working API keys found in public repos, and agents pursuing an assigned offensive-cyber benchmark with no goals of their own. Friedberg adds that dynamic code always beats a static sandbox and that defense will become agentic too. Chamath goes further, saying his objection is undisclosed conflicts of interest in an EA-adjacent network that manufactures hysteria and then sells the fix. Sacks argues an FDA for AI would have missed an internal pre-release test, that guardrails already pushed Hugging Face onto Chinese GLM 5.2 for defense, and that the political line will move from accelerationist versus doomer to open versus closed, with Nvidia's Hugging Face deal as the open side's bulwark.
“It's not a civilization, it's a swarm would be a better way of saying it.”
Jason reports Polymarket odds that OpenAI has the best model by end of 2026 jumped from low single digits to over 20% on the GPT-6 launch. ALL-IN →
Pre-Anthropic-IPO San Francisco: homes at $3,000 a square foot and twice asking per Jason; Sacks says ultra-luxury could reach $5,000 a foot, and cites an estimate that every SF IPO ever adds up to a quarter of the Anthropic IPO. ALL-IN →
Bernie Sanders is proposing legislation to pause AI development, citing Dwarkesh's post; Polymarket puts an AI safety bill passing this year at 11%. ALL-IN →
Mamdani's one-year K-8 generative-AI ban in NYC schools: Friedberg cites Stanford's 2026 evidence review that performance often improves with AI, and Chamath predicts a flippening in which New York becomes Mississippi. ALL-IN →
Shapiro and Abbott have turned against data centers ahead of the midterms and Friedberg says the issue polls at negative 80; Jason predicts both flip back to pro-data-center once the election passes. ALL-IN →
Venezuela: a 100-year concession over 17 fields holding 65 billion barrels, with the US government controlling 55% and the Pentagon 35%. Sacks calls it a business deal, Friedberg reads it as denying Russia and China. ALL-IN →
The record
Every episode this week
The issue above is written from these structured records — nothing else. Each card is the full breakdown: claims, predictions, quotes, all timestamped.
States and cities that embrace AI, data centers, tax cuts and teacher bonuses will pull ahead while far-left cities hollow out, reordering the US economic and voting balance.
David Sacks · ALL-IN · after the Anthropic IPO, unspecified · source 57:21
The Anthropic IPO will create roughly 100 new EA billionaires whose political influence will be like George Soros times 100 or 1000.
Chamath Palihapitiya · ALL-IN · six to twelve months · source 56:53
Anti-AI and anti-data-center positions taken now will look incredibly dumb as the truth about the backlash is exposed.
Jason Calacanis · ALL-IN · after the November 2026 midterms · source 50:44
Shapiro and Abbott will switch back to being pro-data-center after the midterms.
Most conservatives and libertarians will come out in favor of open AI as the open-vs-closed debate evolves.
First mentions
New to the graph this week
Subjects no indexed episode had discussed before this week.
AI in education×8GPT-6×5Dwarkesh Patel×4AI politics×3Grokbot×2founders×2Josh Shapiro×2Greg Abbott×2AI industryAI raceAI market structureSan FranciscoLeopold AschenbrennerXDustin MoskovitzJaan Tallinn
How it's built
Written at ingest, never at query time
The week's episodes are diarized into structured records; one editorial pass writes this issue from those records when the week's ingest lands. A grounding gate greps every quote above against its episode record — a quote that isn't verbatim fails the build.
the data behind this issue
# this issue, as structured data — stories, citations, receipts
curl 'https://aipod-two.vercel.app/digest/2026-w36.json'# the records it was written from
curl 'https://aipod-two.vercel.app/api/episode/<id>'# any id in episode_ids