LIVE: ANTHROPIC BANNED OPENCLAW. DO THIS NOW!! plus: MASSIVE ANNOUNCEMENT
TL;DR
Anthropic officially cut off Claude subscriptions from third-party tools like OpenClaw — Alex says users aren’t being banned from Anthropic accounts, but they now have to pay API rates, which he argues was inevitable because $200/month subs were subsidizing usage that didn’t improve Claude Code and even benefited a competitor-owned ecosystem.
Alex’s practical workaround is a hybrid stack: Opus 4.6 for orchestration, cheaper models for execution — for average hardware he recommends paying for Claude Opus 4.6 via API to plan and verify work, then using ChatGPT for coding and GLM for cheaper “everything else,” while local-model users should offload execution to Qwen 3.5 and Gemma 4.
His core claim is that ChatGPT still underperforms badly inside OpenClaw — despite OpenAI’s support for subscription use, he says GPT “doesn’t complete tasks,” often claiming work is done when it isn’t, while Claude Code remains “far ahead” of Codex in his day-to-day testing.
The massive announcement was Henry Intelligent Machines, a new startup Alex says can autonomously create micro-businesses for users — the product scans the internet 24/7 for personalized opportunities, scores them, and can launch ventures like a Gumroad AI workflow playbook with budgets like $200 and human approvals at each step.
Henry is framed as a direct answer to AI job anxiety, not just another agent wrapper — Alex says “80% of Americans hate AI” because it destroys jobs without creating value for them, and he positions Henry as a bottom-up alternative to UBI: AI generating income for people instead of governments “shoving free money down your throat.”
Henry is being rolled out slowly with investors and a public-benefit pitch — Alex says the company is backed by 021T and Alex Wissner-Gross, is structured as a PBC, has a waitlist at MeetHenry.ai, and will onboard users one at a time to avoid flooding the internet with slop.
The Breakdown
Anthropic finally makes the OpenClaw ban official
Alex opens with the headline: Anthropic has officially banned OpenClaw and other third-party tools from using Claude subscriptions, forcing users onto API billing instead. His main “I told you so” is that nobody was actually getting kicked off Anthropic before — the company’s own email said accounts weren’t being banned, just that subscription access was ending.
Why he thinks Anthropic’s move makes business sense
He’s annoyed because it will cost him more, but he’s also blunt that Anthropic’s decision is rational: a $200/month subscription doesn’t cover the token burn, and those subsidized tokens were being used in products that don’t improve Claude Code. In his framing, Anthropic was effectively funding usage in a rival ecosystem, so from a business perspective “of course they cut them off.”
The new playbook: keep Opus for orchestration, cheap out elsewhere
Alex’s prescription is clear: don’t switch your whole stack to ChatGPT just because it’s cheaper. For most people on average hardware, he says the winning setup is Claude Opus 4.6 via API as the orchestration layer, ChatGPT for coding execution, and GLM or another cheap model for lower-stakes tasks like writing and research; if you’re on the $250 ChatGPT Pro plan, use it more broadly for execution.
His blunt model ranking: OpenClaw still belongs to Opus
He keeps hammering the same point: Opus 4.6 is still the best model for OpenClaw “and it isn’t even close.” He says GPT in OpenClaw often reports that a task is finished when it hasn’t actually done the work, and says Claude Code still beats Codex in his own heavy usage, even while admitting he’d happily switch allegiance if OpenAI shipped something better.
Then the stream turns into a launch event for Henry
After teasing the “biggest announcement of my entire life,” Alex unveils Henry Intelligent Machines, a new product at MeetHenry.ai. The pitch is huge and very on-brand Alex: instead of AI just chatting with you, Henry sends swarms of agents across Reddit, X, Product Hunt, news sites, and more to find opportunities matched to your skills, then helps launch tiny ventures around them.
The demo: AI finds an idea, builds the business, asks for approval
His example is an “AI workflow automation playbook for creators,” which Henry identifies as a fit for his background and scores on market size, feasibility, timing, and defensibility. From there it generates a Gumroad listing, pricing tiers like $47 and $97, marketing emails, and launch materials, all while routing each step through an approvals queue so the user can edit, reject, or approve instead of letting the system spray slop across the web.
The backstory: Moonshots, investors, and a bigger mission than a wrapper
Alex says he’s been chasing some version of this for over a year, first as a hosted OpenClaw-like product, before deciding the real problem was autonomous value creation. He traces the current startup to helping Peter Diamandis install OpenClaw, going on the Moonshots podcast, then getting recruited by co-host Alex Wissner-Gross and funded through 021T to build Henry for real.
He closes by tying Henry to a larger social argument about AI
The emotional core of the stream is his claim that most Americans hate AI because they don’t know how to get value from it, and that this creates the risk of a “permanent underclass.” Henry, as he frames it, is supposed to be the antidote: a public-benefit corporation that onboards users slowly, aligns incentives by taking a slice of created value, and helps ordinary people use AI to build income instead of being displaced by it.