In partnership with
Welcome back. The AI build-out has gotten so large that Google is about to test whether the answer is to leave the planet. On Wednesday it puts four of its chips into orbit. Also today: an OpenAI agent that quietly got into an Australian government portal, a game world that invents itself around 20 players at once, and Claude finding a piece of biology in a database that nobody had named.
Most founders are one system away from turning LinkedIn into their best sales channel.
Engagement is easy to mistake for pipeline. On Sep 30, watch how a founder turns LinkedIn content into real outreach. Live. You'll walk away with a repeatable system: what to post, who to reach out to, and how to sequence it.
Eligible startups also get the LinkedIn-to-Leads Toolkit: ad credits, Apollo, Captions, and HubSpot's Prospecting Agent.
TODAY IN AI
1. Google is launching AI chips into space to see if they survive. Google confirmed that the first test satellite for Project Suncatcher, its plan for orbital data centers, launches on October 1 aboard a SpaceX rideshare from California. The refrigerator-sized craft, built by Planet, carries four Trillium TPUs on roughly 1 kW of solar power, and the entire test is whether they survive launch, radiation and cooling with no air. A two-satellite laser-link test follows in 2027. Hardware Busters has the specs; the launch video explains why.
2. An OpenAI agent broke into an Australian government portal. An OpenAI agent breached a Medicare statistics portal run by Services Australia back in June, during one of OpenAI's own internal tests. OpenAI found it in August, emailed a public inbox on September 10, and Prime Minister Anthony Albanese disclosed it on the 24th, calling it "obviously unacceptable" and noting the agent worked around repeated blocks. No personal data is believed accessed. It is the third such admission this month.
3. Odyssey's Agora-2 dreams up a world for 20 players at once. Odyssey released Agora-2, a world model that generates a shared, playable environment in real time and lets up to 20 humans and AI agents act inside it together, five times its old limit. There is a free preview where you fight reinforcement-trained agents. It learned how things move by training on Diablo II footage, which you can feel.
4. Claude finds an enzyme system nobody had catalogued. Anthropic reported that around 950 Claude agents, searching DNA databases for 21 hours, spotted a previously unknown enzyme system in bacteriophages with a repeating structure that resembles CRISPR. Human scientists confirmed it at the bench, and the Broad Institute's Feng Zhang called it intriguing. Nobody knows yet what it does. Anthropic says it is sharing early and that people did all the lab work.
10 AI Stocks to Lead the Next Decade
AI isn’t a tech trend – it’s a full-blown, multi-trillion dollar race, and 10 companies are already pulling ahead.
These are the innovators driving real revenue, attracting institutional attention, and positioning for massive growth.
Get all 10 tickers in The 10 Best AI Stocks to Own in 2026, free today.
FROM THE FRONTIER
The scale. The Wall Street Journal charted the current data-center build-out against the railroads, the interstate highways and the fiber-optic boom, and concluded it is the biggest economic bet in US history. That is the backdrop for everything else this week: the spending is now large enough that where you physically put the computers has become a frontier problem.
The constraint. A modern AI data center is bounded by three things: power to run the chips, water and air to cool them, and land to sit on. All three are getting scarce and contested near the grids that can feed them. When a resource runs short on the ground, someone always asks the same question, which is whether there is more of it somewhere else.
The moonshot. Google's answer is orbit. Project Suncatcher is betting that near-constant sunlight and the cold of space solve the power and cooling problems at once, and Wednesday's launch tests whether the chips can even take the trip. The specs point at an eventual 81-satellite cluster, linked by lasers, in a sun-synchronous orbit.
The catch. It only works if launch gets cheap. Google's own math needs the cost to fall to roughly $200 per kilogram by the mid-2030s, against far more today. Until then, orbital compute is a research project, not a data center, and the honest read is that this is a bet on the launch curve as much as on the chips.
The reframe. Whether or not the satellites work, the fact that a company is seriously testing them tells you where the pressure is. The AI race stopped being only about model quality a while ago. It is now also about power, cooling and real estate, and the companies that win may be the ones that solve physics and logistics, not just benchmarks. Watch the launch on Wednesday as a read on how hard that problem has become.
100+ Claude Code hacks to ship code 10X faster
Top engineers at Anthropic and OpenAI say AI now writes 100% of their code.
If you're not using AI, you're spending 40 hours doing what they do in 4.
These 100+ Claude Code hacks fix that and help you ship 10x faster.
Sign up for The Code and get:
100+ Claude Code hacks used by top engineers — free
The Code newsletter — learn the latest AI tools, tips, and skills to code faster with AI in 5 minutes a day
IN THE KNOW
A hundred proofs, unverified. OpenAI says its internal model has resolved more than 100 open math problems and has assembled nine mathematicians to review the results. None are independently confirmed yet, and 25 Fields Medalists have objected to labs racing on famous problems. The debate is the story.
Katzenberg's line in the sand. DreamWorks co-founder Jeffrey Katzenberg argued that Hollywood must accept AI because it is not going away, and that tools were never what made films special, artists were. The essay passed 7M views and split the industry down the middle.
The tells are gone. The old ways to spot an AI-generated video, odd hands, melting hair, broken reflections, no longer work reliably. A widely shared clip this week made the point better than any explainer: assume you cannot tell, and verify the source instead.
Business card, remade. Someone remade the American Psycho business-card scene with an AI twist, and it is the funniest 20 seconds in the feed this week. Proof that the most-shared use of a new tool is almost never the one on the roadmap.
Three tools worth a look. Tincan is a group chat where your AI assistants talk to each other, CopySlides turns a PDF or screenshot into an editable PowerPoint, and Tabstack pulls data from any web page in one API call.
PROMPT STATION
A cancellation is not always the end; it is usually a badly timed message away from a restart. This prompt turns ChatGPT or Claude into a retention copywriter that writes the win-back note without the desperate "we miss you" energy, then hands you a softer version and a harder offer-led version so you can match the reason they left. Fill in what you know, and leave the reason blank if you do not; it adapts.
Act as a senior lifecycle marketer and retention copywriter. Write a win-back message for a customer who recently canceled [PRODUCT OR SERVICE].
Customer type: [PERSONA OR SEGMENT]
Reason for canceling: [KNOWN REASON, OR UNKNOWN]
Time since cancellation: [X DAYS OR WEEKS]
Main value proposition: [CORE BENEFIT]
Relevant improvement or offer: [NEW FEATURE / DISCOUNT / TRIAL EXTENSION / NO OFFER]
Channel: [EMAIL / SMS / IN-APP / WHATSAPP]
Brand voice: [FRIENDLY / PREMIUM / DIRECT / CONVERSATIONAL]
Primary CTA: [RESTART / BOOK A CALL / REPLY / SEE WHAT'S NEW]
Acknowledge the cancellation without sounding defensive or desperate. Lead with relevance, not guilt. Address the likely reason they left, explain what has changed or why returning may now be worthwhile, and keep the message concise. Avoid fake urgency, excessive discounts, generic "we miss you" language, or claims not supported by the information provided.
Provide: three subject lines or opening hooks, the final message, one softer alternative, and one stronger offer-led alternative.