Good morning. It's Tuesday, November 3, and we're starting with a look inside how we built an AI second brain that manages work, remembers commitments, and gets things done before anyone even asks.

Then: Anthropic warns self-improving AI may be imminent, China sets its sights on a crewed Moon landing, AI demand strains PC supply chains, and OpenAI prepares its post-election regulatory agenda.

Plus: a poll on AI's potential to influence the 2026 midterm elections.

What if your AI didn't just answer questions, but remembered your commitments, caught mistakes, prepared for meetings, and started work before you asked?

Over the past few months, we've built a system that connects email, calendars, CRM data, documents, and business metrics into something closer to an AI chief of staff. It's already saving an estimated hour a day, but getting it to work reliably has been anything but straightforward.

In this hands-on guide, we break down how the system was built, the tools and workflows behind it, what went wrong, and how you can start building your own.

→ See how we built our AI second brain

Chinese space officials released their clearest schedule yet for landing astronauts on the Moon before 2030. The plan calls for an uncrewed Mengzhou spacecraft test in early 2027, a lunar-lander test in 2028 and a full lunar-orbit rehearsal in 2029. The first test will show whether the schedule can hold as China races NASA back to the lunar surface.

Anthropic warned in an updated IPO filing that it may be approaching recursive self-improvement (RSI), the point where AI systems accelerate the development of increasingly capable AI. The company acknowledged that AI-assisted research has already produced capability jumps it failed to recognize immediately, raising concerns that progress could outpace safety measures.

OpenAI is testing one ChatGPT interface that moves from conversation to action without making users switch between Chat and Work tabs. Files, plugins and tasks such as building websites or presentations will live together, while Codex remains separate. ChatGPT and Codex will also draw from one token allowance, which could make some heavy users hit their limits sooner.

Some Grok Bot users saw their weekly limits jump to 100% even when they had not been using the agent. Elon Musk said serving the product had become “unsustainably high” relative to revenue, but he has not explained the new limits. The next move on pricing or usage caps will test whether expensive autonomous agents can support mass-market subscriptions.

AI data centers have already absorbed much of the world’s advanced memory and graphics-chip supply. Now the squeeze has reached printed circuit boards: factories are prioritizing complex, higher-priced boards for AI servers, while HP, Dell and other PC makers worry that ordinary motherboard orders will arrive late. One Taiwanese manufacturer said laminate prices are at least 12 times their July level, and some new board orders may not ship until late 2028.

Circuit boards sit underneath nearly every electronic product, so the shortage can raise prices and delay PCs, cameras and medical equipment even when the chips themselves are available. Suppliers plan to add capacity, but skilled-worker shortages and long construction timelines limit how quickly they can respond; watch whether promised expansion by 2028 catches up with AI demand. → Read the full article here.

OpenAI is preparing a policy and public-relations push immediately after the U.S. midterm elections, according to Axios. Its proposals include mandatory “speed limits” for the most powerful AI systems, community agreements for new data centers, worker support and an effort to rebrand data centers as “AI compute campuses.” The campaign is expected to begin during Congress’s lame-duck session on November 9.

The plan is an attempt to shape the rules before a potentially less friendly Congress writes them. OpenAI is promising more human control and local accountability while also trying to preserve room for rapid expansion; watch which ideas become actual bills or enforceable commitments rather than messaging. → Read the full article here.

AI-generated images, videos, and fake news are becoming increasingly difficult to distinguish from reality. As these tools become more accessible, how concerned are you about their potential to mislead voters and influence election outcomes?

Hugging Face Biology Research released Carbon-X, an open 14.8-billion-parameter model trained on both DNA sequences and ordinary text. Researchers can paste in genetic code and ask questions about mutation effects, regulatory regions or likely next bases; its creators report a 66.7% average across five DNA benchmarks, close to GPT-6.1 Sol’s 67.5% score at far lower estimated inference cost. → Explore Carbon-X.

Apple aims for a $1,000 home robot: Roaming robot targeted for 2027 would handle FaceTime and lights, priced at $1,000 or more.

Claude breaks the coding interview: Anthropic's Felix Rieseberg dropped coding puzzles from interviews after Claude outperformed most candidates.

South Korea funds a sovereign AI push: Proposed 2027 budget: 21.3T won for chips, physical AI, data centers, plus 4.7T for a sovereign model.

Zest books restaurants by text: Text concierge recommends restaurants, checks availability and books tables via OpenTable, Resy and SevenRooms.

Adobe gives editing agents specialist tools: Adobe and UCSD's EditAgent mixes general and specialist editing tools, guided by measured image data.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team