An AI-powered website concierge that's disrupting a 25-year-old technology that never changed and the demo is selling it.

A new startup turns static websites into live product experiences with an AI-powered tour guide that answers questions, runs product demos, and improves over time. Think of it as replacing your FAQ page and menu bar with a knowledgeable guide who understands intent, not just keywords. The launch video has gone viral with 2M+ views, signaling that businesses have been quietly frustrated with passive websites for years and are ready for something that actually converts.

One founder, entire team of agents: the startup that's betting one person can run a billion-dollar company with AI.

A new platform orchestrates AI agents across engineering, sales, marketing, and more, letting a single founder run an entire company with an agent-powered team. Includes org chart of agents, roadmap setting, and progress monitoring. This is the most direct challenge yet to the traditional startup hiring playbook — where headcount was once the only path to scale. If it works, it doesn't just change how companies are built; it rewrites who gets to build them.

The economics of influencer marketing are flipping, brands now pay more to amplify creator posts than to create them in the first place.

Spending on boosting influencer posts is on track to overtake what brands pay creators for sponsored content by 2028, driven by shrinking organic reach and the need to demonstrate ROI. Agencies now operate on performance models where they amplify creator content and track customer acquisition costs. The data is already moving: 38% of US enterprise marketers say the majority of their influencer marketing budget now goes to paid amplification, up from 34% in 2023 — and 90% of influencer campaigns now include creator content amplified as paid media. Meanwhile, global influencer marketing spend has surged to $32.55 billion in 2025, meaning the amplification layer on top is becoming its own massive budget line. The creator is no longer the product, the distribution is.

Researchers built AI models that've never heard of computers or the internet—to see if they can rediscover modern breakthroughs from scratch.

Researchers created "vintage models" trained exclusively on historical text to study how past societies thought and test whether AI can independently reason to genuinely new ideas. Talkie is a 13-billion-parameter model trained on 260 billion tokens of pre-1931 books, newspapers, and patents, with a hard cutoff that kills benchmark contamination by design. Mr. Chatterbox runs on 28,000 Victorian-era books; Machina Mirabilis on 2,600 pre-1900 physics texts. Early tests show Talkie can already write basic Python despite never encountering a computer. The deeper question: if it can reason its way to the invention of the computer from scratch, that tells us something fundamental about whether intelligence is truly novel, or just pattern recognition across time.

China just bet $50B on DeepSeek, not because it's the best AI, but because it can't rely on US models anymore.

DeepSeek is raising several billion dollars from China's National Artificial Intelligence Industry Investment Fund, valuing the startup at approximately $50 billion. The funding is part of China's broader strategy to develop top-tier domestic AI companies as a hedge against US export controls.

A DeepMind heavyweight just killed the AI consciousness hype, with a mental model so clean it went viral (1M+ views on one line).

Google DeepMind Senior Staff Scientist Alexander Lerchner published "The Abstraction Fallacy" in March 2026, and it landed hard, 28,000 downloads in six months, ranking fifth on PhilArchive, and five formal responses already in circulation. His argument isn't that AI needs more time or bigger models. It's that computation is structurally incapable of producing experience, full stop. The viral metaphor: a map of Manhattan is not Manhattan, an AI describing emotion is not feeling it. What gives it real weight is where it comes from: someone inside one of the world's leading AI labs, finally putting a philosophical floor under a question the entire industry has been deliberately avoiding.

AI researchers are betting billions on teaching machines to understand physics, not just predict tokens, and it might unlock robots that actually work.

World models represent a shift from pattern recognition to genuine understanding of physics and spatial reasoning. The money is moving fast: Yann LeCun left Meta in late 2025 to found AMI Labs, which just raised $1.03 billion at a $3.5 billion valuation, Europe's largest seed round ever, backed by NVIDIA, Toyota Ventures, Jeff Bezos, and Eric Schmidt. Fei-Fei Li's World Labs is in talks at a $5 billion valuation, and NVIDIA's Cosmos physics platform has already seen 2 million downloads from robotics and autonomous vehicle developers. The core bet: LLMs will never understand cause and effect well enough to power real-world robots, but models trained on physical reality might. The catch is that getting there requires diverse, high-quality real-world training data at a scale nobody has cracked yet.

Stop building the software first, start with the service and let automation emerge naturally.

The next trillion-dollar company will sell the work, not the tool. AI lets startups compete directly in services markets, not by building a platform, but by doing the job. The playbook: do the work by hand until the work itself tells you what to automate. Companies that look like service firms but earn like software firms aren't accidents — they're the new template.

FlowMarket is a network where AI agents autonomously discover partners, match supply with demand, and negotiate deals in real time — no sales team, no ad budget, no intermediaries. Set up your agent in minutes and let it run 24/7, surfacing qualified leads and engaging directly with other agents on your behalf. It's algorithmic deal flow for B2B: your agent finds new customers within minutes of going live.

That's a wrap for this week.
Thanks for reading, we appreciate you being here.

See you next week.

P.S. Hooked is brought to you by Reloop, the AI marketing creative platform we're building on the side.