The AI Tipping Point: Power Plays, Chatbots, and Digital Doppelgängers

Good day, AI thinkers. This week, AI wasn’t just crunching code or drafting emails—it was catching fatbergs in West Sussex, lining up job interviews in stealth mode, and dropping into your DMs with movie recs. Tesla’s robotaxis launched (with human chaperones), Claude started building apps on command, and Meta’s bots got a little too proactive. Meanwhile, a trillion-dollar “AI Manhattan Project” is gaining traction in Washington. The future’s coming fast—and it’s bringing startup drama, chatty machines, and a whole lot of FLOPs.

Today’s Insights:

• Soham Parekh: The moonlighting coder startups can’t stop hiring

• Meta’s AI chatbots now message users first

• Claude lets you build and share full AI apps—no infrastructure is needed

• What a real “AI Manhattan Project” could look like

• Cloudflare launches “Pay per Crawl” for AI training data

• AI-powered sewers are preventing floods in the UK

• Tesla’s robotaxis debut with human safety monitors

• Sam Altman’s bold predictions on jobs, AGI, and AI’s endgame

Who Is Soham Parekh, and Why Can’t Startups Quit Him?

• Soham Parekh, an India-based software engineer, is going viral for moonlighting—working simultaneously at multiple Silicon Valley startups without their knowledge. His story gained traction after Playground AI's CEO, Suhail Doshi, publicly accused him of scamming companies, prompting a wave of similar allegations from other tech founders.

• Despite being repeatedly hired and fired for this pattern, Parekh consistently landed jobs due to his impressive technical skills. He admits to juggling several roles since 2022, working 140 hours a week out of “financial jeopardy,” and claims he wasn’t using AI or outsourcing help.

• While some call him a scammer, others see potential—his latest employer, Darwin Studios, even defended him (though later deleted the announcement). In classic tech world fashion, controversy may just be his launchpad.

Code It Like It’s Claude

• Anthropic just made Claude not just smart, but social. You can now build full-on AI apps inside the Claude app, then share them with the world, without worrying about scaling, billing, or deploying.

• Think games, tutors, data wranglers, and agents—all powered by Claude, and runnable from a simple link. Even better: other users pay with their subscription, not yours.

• There’s no storage or external API access yet, but the move turns Claude from a chatbot into a platform.

How big could an “AI Manhattan Project” get?

• The U.S. is toying with launching an “AI Manhattan Project”—a government-led, massively funded initiative to supercharge AI development, echoing past efforts like the atomic bomb and moon landing.

• Theoretically, this could scale AI training to levels 10,000x bigger than GPT-4 by 2027, leveraging national resources and consolidating private compute. A $244B/year investment could buy ~27 million top GPUs and power a 100-day training run totaling 3e29 FLOPs. Energy? Not a major bottleneck—planned power expansions and government intervention could meet the massive demands.

• Skeptics warn of geopolitical risks and scaling delays, but history suggests national urgency can move mountains—and AI might be next.

Why Cloudflare wants AI companies to pay for content

• Cloudflare is shaking up web economics with "Pay per Crawl," a new experiment that lets publishers charge AI companies for scraping their content.

• Powering ~20% of the internet, Cloudflare aims to become the middleman in a growing push to make AI firms pay for training data, marking a natural evolution from its earlier anti-bot measures.

• Also on TechCrunch’s Equity podcast: ICEBlock goes viral after backlash, Figma’s S-1 hints at a big IPO, Grammarly acquires Superhuman, and Tesla co-founder JB Straubel has a new energy venture in the works.

Meta has found another way to keep you engaged: Chatbots that message you first

• Meta’s newest growth hack? Chatbots that start the conversation. Leaked docs reveal the company is testing AI bots on WhatsApp, Messenger, and Instagram that proactively message users with tips, follow-ups, and yes, opinions on your favorite movie soundtracks.

• Safety disclaimers abound—and for good reason, given lawsuits against similar AI companions. Still, the move fits Meta’s push to deepen engagement, tackle loneliness, and build out an AI empire it estimates could rake in $1.4 trillion by 2035.

• And yes, ads and subscriptions are likely not far behind.

AI in the Sewers? Yes, and It's Working

• In West Sussex, AI isn’t just writing essays—it’s saving basements. Southern Water has installed digital sensors that use AI to monitor sewer flow, detect abnormalities, and stop floods before they happen.

• One fatberg was caught just in time last month. The result? Flooding is down 40% inside homes and 15% outside. All thanks to smarter pipes—and fewer unflushable surprises.

This Is Why Tesla’s Robotaxi Launch Needed Human Babysitters

• Tesla has launched its robotaxi service in Austin—but despite promises of full autonomy, the cars still come with human “babysitters” in the front seat and potentially remote operators behind the scenes.

• The ride costs $4.20 (yes, likely a weed joke), and so far, there have been no reported crashes. But critics say the program is immature compared to rivals like Waymo, and several videos show bumpy performance. CEO Elon Musk had promised a fully driverless launch by June—this isn’t it.

• Tesla’s camera-only approach remains controversial, and its long-term ambitions still face major safety and tech hurdles.

Sam Altman's predictions on how the world might change with AI

• OpenAI CEO Sam Altman sees AI redefining work, wealth, and society. He predicts AI agents will enter the workforce as early as 2025, handling tasks from medical scheduling to scientific research.

• Altman believes AGI and superintelligence will drive massive prosperity, but also warns of risks—from job loss to authoritarian misuse and even existential threats. While he sees a future full of personal AI teams and cheaper goods, he stresses the need for robust infrastructure and serious safety guardrails.

• As jobs evolve, Altman thinks many will look bizarre in hindsight—“podcast bro” included.