Anthropic says AI is improving itself, tokenmaxxing is over, AI is coming to your personal life, Supabase doubles valuation

In this episode of Get the Check, the pod breaks down Anthropic's RSI blog, the end of the tokenmaxxing era, whether personal AI assistants are finally close, and Supabase hitting a $10B valuation because of the rise of vibe coding.

They start with Anthropic's recursive self-improvement paper, which says 80% of code merged at Anthropic is written by Claude, and that Mythos makes the right research decision 64% of the time. The hosts get into what RSI actually means, is it AI improving its own weights, or just better scaffolding around the same model? Maya argues we haven't seen AI solve a proof or generate a genuinely new hypothesis that a human didn't think of first. Priya thinks lines of code is a terrible productivity metric, but also the only one people can agree on. The consensus: Anthropic isn't saying anything that new, and the Wall Street Journal's "Anthropic pauses AI" headline was rage bait (at least for Maya..)

Next, the tokenmaxxing era is officially over. Uber blew through its entire 2026 AI budget and it’s only June. Cognition announced it will pay companies up to $10 million if Devin doesn't deliver ROI, which is a bold move that makes Cognition the benchmark-setter for its own product. And app-layer companies are quietly moving to open-source models like Qwen to cut spend by 80-90%, which Maya thinks reverses the minute Anthropic drops Mythos, but Anika has a different take

Then there’s Bernie Sanders' American AI Sovereign Wealth Fund Act, which would require OpenAI and Anthropic to hand 50% of their equity to the federal government. His argument is that these models trained on everyone's data without consent. Somehow both sides, the AI companies and politicians, kind of agree in principle but disagree on the mechanism. Maya's take: if these companies just went public, wouldn't that solve it, but once again Anika has a different take lol.

Finally, the pod digs into the personal AI assistant moment - why Town AI ($55M Series A from a16z) and Poke's Apple Business messaging debut both launched this week. The category has taken so long despite being the "obvious" use case.

Finally, Supabase hitting $10B, 40% of recent YC batches are using it, their Lovable partnership auto-provisions databases for vibe-coders, and Maya explains the Firebase migration issues that Supabase was built to fix.

This episode is brought to you by:

1. Kalshi — Anthropic's IPO odds are live on Kalshi right now. Sign up at kalshi.com/sign-up?referral=getthecheck and get $10 when you trade using our code.
2. Known — Use code GETTHECHECK at known.com to skip the waitlist and get a free date.

Follow the pod on Instagram and X @getthecheckpod.

Chapters:

00:00 Maya forgot her pre-planned banter (classic)

00:17 Priya found Maya the perfect bag in Singapore

04:05 Anthropic says AI is building itself, but is that actually true?

07:22 Wait can we define RSI please

14:57 Tokenmaxxing is officially over

15:23 Cognition's $10M AI productivity guarantee

19:18 App-layer companies are quietly switching to Qwen

22:59 The WSJ's Anthropic headline was rage bait for Maya

24:24 Bernie Sanders wants to seize 50% of OpenAI and Sam Altman is down to talk about it

26:05 Maya thinks going public already redistributes AI wealth

31:56 The personal AI assistant is still not here

36:08 What actually changed with Poke's Apple Business launch

39:22 Town AI raises $55M

49:45 Supabase hit a $10B valuation