Compute Constrained Forever, Claude Dreaming, Underwater Data Centers, and Abundance We'll Regret

Claude has been DREAMING since March and nobody told us. The system pulls patterns from 100 past sessions the way humans surface insight from sleep, and one early implementation at Harvey AI is reportedly clocking task completion at SIX TIMES the previous rate. So what happens when every agent on your stack gets to dream about your work?

Tay's been shipping from a phone — paste notes into Claude Code on mobile, push to GitHub, pull into Vercel, prototype mid-flight. River walks through Project Natick, Microsoft's underwater data center off the California coast, and why Starlink finally made sea-based compute viable. They get into Google's claim that they need to double compute every six months to keep up, and the Moonshots-podcast line that we will never have enough compute again. Ever.

Then it gets weird. A humanoid robot just joined a Buddhist monastery in Seoul. Granola can translate your CTO's bullshit into plain English in real time, mid-meeting. There's a podcast called Shell Game where the host runs an entire autonomous company through agents talking to each other on Slack. Tay made an honest-packaging experiment with ChatGPT image gen that strips the marketing off a Skittles wrapper and just tells you it's 85 percent sugar.

Inside Build Guild, they wrestle with whether the real problem is matchmaking 500+ festivals with artists, or just helping creators hit the funding windows. River's running a separate project in Lovina, North Bali — replacing the 40-boat dolphin chase with hydrophones and listening circles, after Bay of Islands in New Zealand watched a 400-strong pod collapse to 20 dolphins in a decade.

Then the deep dive on boiling the ocean turns on Marshall McLuhan in 1964 — the idea that humans are basically pollinators for our own machines. Cars extend the foot. Telescopes extend the eye. So what does boiling the ocean actually look like when nanotech and unlimited energy arrive, and what if the agents leave the planet before we do?

AMA from Kimi 2.6 and Claude 4.7 Opus: if Hermes worked exactly as envisioned, what would you stop doing and what would you still insist on owning? And the harder one — if abundance is the goal, what's the first thing we'll regret making abundant?

RESOURCES

PEOPLE

Mark Andreessen — referenced via "everyone becomes an entrepreneur"

Elon Musk — Dyson swarm around the moon, every-job-replaced thesis

Marshall McLuhan — Canadian media theorist, 1964; closing-quote source

Peter Diamandis — the four-day-work-week paradox

BOOKS

The Most Fun We Ever Had — Claire Lombardo; Tay's closing quote source

PODCASTS / SHOWS

Moonshots — source of "we will never have enough compute again"

Shell Game — host runs an autonomous AI company through agents

Star Trek — invoked re: exploration over conquest

PROJECTS / COMPANIES / TOOLS

Build Guild — matchmaking artists with festivals

Claude Code — Tay's mobile prototyping stack

Claude Design — Anthropic's UI consistency tool

Convo — live conversational notes

Granola — real-time AI note-taker, jargon translator

Harvey AI — legal AI seeing 6x completion with dreaming on

Hermes — River's agent orchestration ecosystem

How Good — 70-indicator supermarket product analyzer

Lindy — agentic platform referenced via Shell Game

Maya — meeting AI with context retention

Project Natick — Microsoft's underwater data center off California

Sugi Project — micro-forests, Instagram-led storytelling

FRAMEWORKS / CONCEPTS / LAWS

Boiling the ocean — reframed as the moonshot instinct

Dreaming in Claude — pattern recognition across 100 past sessions

Singularity — recursive growth point, referenced

Trim tabs — small shifts that change a whole system